When You Can't Show the Result, Show Being Understood

A small cosmetic-dental clinic had an offer, no brand guide, and advertising rules that took the before-and-after off the table. Four short films, made entirely in code and passed by independent judges, show what a code-made film series looks like for a small business, and how the constraint pushed it to its strongest idea.

Scott Farrell · LeverageAI · September 2026

A phone photo from a friend's fortieth, on a terrace by the beach. A woman pinches in on her own laugh until one tooth fills the screen.

"Every photo… it's the first thing I look at. It's always been a bit shorter than the others."

She circles it in gold.

Sixty-five seconds later the film ends on the same photo, unchanged, pulled back until you can see the whole table of friends. There's no reveal. No new smile, no mirror moment, no "after". Nothing about her has changed.

It's an ad for cosmetic dental work that never shows a result. It's the main film of a series of four I made this week for a small cosmetic-dental clinic. The clinic had an offer to promote: two spring composite-veneer packages. It didn't have a brand guide, a written brand promise or a marketing person, and the rules for advertising this kind of treatment in Australia took away the move every veneer ad leans on. Nothing in the films was shot on a camera. The story, the look, the voices, the music and the review were all code.

Most writing about AI-made marketing shows you the output and hides the loop. This is the other way round. It's the whole specimen, opened up: the brief, what each film is for, how each one tells its story, what the judges said and what they made us change, the checklist that has to be worked through before any of it is posted, the number that will say whether it worked, and what the making left behind for the next business. It also covers what none of it proves yet, which is more than you'd think.

A clinic with an offer, and nothing else

Here's what the clinic had. An offer page with two packages on it, one where the dentist shapes the composite by hand and one where a clear mould carries a planned shape onto the teeth. Its own colours, its own type, its own package cards. A real enquiry form. A treating dentist with a photo and a registration line. And that's it.

That's the normal case for a small practice, and it isn't a criticism. The people running it are good at dentistry. Nobody has ever asked them to write down who their customer is, what they promise that customer, or how they sound. When I looked back on the job, that's the thing that struck me:

"Small businesses generally don't have an experienced professional running their marketing. Ask them what's your brand guide, what's your language, what's your brand promise, and they wouldn't know. It's absent."Me, after the films were finished

The conventional creative workflow starts from a brief the client supplies. Here there wasn't one to supply, and asking the owner to fill in a brand questionnaire would have been asking them to do the professional work they needed help with. So I wrote the brief myself, before any film existed, generated with AI from my own frameworks and what we'd already learned making films for other businesses. It carried a central question, "What would you change, and what would you keep?", and an instruction I'd want in every brief for a service business: no "we understand your needs" narration. Let the viewer watch someone's needs becoming understood.

It also labelled its own guesses about the audience honestly, as "creative hypotheses to test, not findings from patient research." I'll come back to where those ideas came from, because the records tell a less tidy story than the finished films do. (The general case, what to do when a client can't brief you, is The Taste Harness's subject. This is the one instance, walked end to end.)

The rule that took the reveal away

Most veneer ads you've scrolled past end the same way. The before, the after, the delighted patient. It's the grammar of the category, and in Australia it's now tightly fenced.

Since September 2025, advertising for higher-risk non-surgical cosmetic procedures has had its own national guidelines, and the first example of such a procedure they name is dental veneers.¹ They're blunt about testimonials: advertising for these procedures "must not use testimonials due to their potential to create unrealistic expectations of beneficial treatment", and that includes "patient stories and experiences" and "purported testimonials".¹ Before-and-after images aren't banned outright, but they "may create unrealistic expectations", and where they're used they must be genuine images of actual patients treated by that practitioner, unedited, with a prominent warning that the outcome is only relevant to that patient.¹

A film made from code and generated stills can't meet that condition. There's no real patient in it, so there's no real result to show. For this series, the reveal was off the table before a frame existed.

The review page we wrote to present the series, one page that explains every film, made an argument I agree with. It said the result shots aren't just forbidden. They'd make weaker films anyway, "because they promise an outcome nobody can promise."

Then there's a detail the research for this article turned up. The regulator publishes example adverts, some marked as problems and some as compliant. One of its compliant examples reads, in part: "During a consultation with our team, we will discuss your treatment goals and options. A personalised treatment plan will be created for you."² So the rules push every clinic towards the same place: stop selling the outcome, talk about the consultation. The regulator hasn't seen these films, and nothing here is a claim about how it would view them. The point is about direction: its example and the films both start from the consultation, with one difference that matters. The compliant example tells you the plan is personalised. The films try to let you watch it happen.

Which left one question to design the whole series around. If you can't show what happens to her teeth, what can you show?

What happens to her words.

People don't arrive with dental words

The review page opened on the idea that answers that question, and it's the best line in the series:

"People don't arrive with dental words. They arrive with one tooth. Nobody books a consultation thinking 'lateral incisor proportion'. They've noticed something in photos for years: one tooth a bit shorter, a chip, an edge worn flat. What they're unsure about is whether it can change without everything changing, and whether they'll be talked into more than they wanted."The series' review page

Read that last sentence again, because the whole series is an answer to it. The worry isn't "will it look good?" It's "will I lose control of this?" Will one tooth become ten? Will I walk in with a small question and walk out with a treatment plan I never asked for?

So the films don't claim that the clinic listens. In the page's words, "They let the viewer watch it happen: someone points at the part they mean, says what they'd change and what they'd keep, and those words visibly carry through into the plan." In the main film, the claim "we listen" becomes a consultation note filling itself in with her own words:

That's the whole mechanism in two lines. It's what I've called a miniature worked example in Demonstrate Over Tell: this happens, the service responds, this is what I get. The page put it in one sentence I'd happily put on the wall of any agency: "A viewer takes on trust what they're told, but they believe what they watch."

The page set out six principles for the series. Each one is about the viewer, not the clinic:

The third one is the device the series turned on, and it's worth one sharpening. My thinking partner, an AI I work through ideas with, put it like this after reading the records: "Their words do not merely reappear. They change what happens. That is the difference between remembering someone's words and demonstrating that they were heard." A film could paste "keep the colour" onto three screens and still show exactly the plan it would have shown anyway. That's theatre. The films have to show a consequence: the colour kept, the whiter shade never offered, the bigger commitment not pushed. The general test for that, and the research on why echoing someone's words back doesn't make them feel understood, is in The Taste Harness. Here, the question is whether each film actually passes it.

Four films, one next step

A small business asking for "some videos for the offer" usually gets one ad in three lengths. This series was planned differently, and the review page says how in the line from the whole job I'd copy first: "Four films, one next step. Each film does one job, for one moment in someone's decision."

Think about who's actually out there. Someone scrolling past with no intention of thinking about their teeth today. Someone who's curious and will watch a minute if it's about them. Someone who's already looked at the two packages and can't tell them apart. And someone who watched, nodded, and still hasn't asked, because asking feels like the first step onto a conveyor belt. Those are four different people, or one person on four different days, and a single film can't serve all of them. So the series gave each moment its own film:

The shared ending is doing more work than it looks. All four films end on the ask the offer page itself makes, to request a discovery call, and the page is explicit that it's small on purpose: "A request isn't a booking, and any consultation fee is explained before anything is booked." That's the right size of ask for people at the start of a decision. Nielsen Norman Group's usability guidance on forms notes that "people in the initial stages of the buying cycle have lower commitment and a higher propensity to abandon forms than people in later stages", and the advice that follows is to tailor what you ask to "the buyer's state and commitment level."³ A booking button asks a hesitant person to commit to treatment. A request for a call asks them to have a conversation. Four films with one small ask means nobody meets a bigger ask than their moment can carry.

Each film comes in two shapes: wide (16:9) for the offer page, YouTube and desktop feeds, and tall (9:16) for phones, "with the words big and clear of the app's buttons." That last clause is a real constraint, not styling. Meta's own guide for tall Reels ads asks you to keep "at least 14% of the top, 35% of the bottom and 6% on each side" free of text and logos, because that's where the app draws its buttons.⁴

The page gave its reason for big words as "most people watch muted first". That's half right, and the half that's wrong matters. In feeds, it's well supported: Meta told advertisers years ago that "most video ads in mobile feed are viewed without sound."⁵ But Reels "default to sound-on".⁶ The safe design is Meta's own advice from the feed era: a video should "communicate clearly without sound, but get better or funnier with sound on."⁷ That's what the films ended up doing, for a reason you'll see in the main film: the offer is on screen for the muted and spoken aloud for everyone else.

Read the four films together and the thing they share isn't the palette or the typeface. It's a relationship. My thinking partner set it out as a table, and I'd use it as the brief for any series:

Its own caveat came attached, and it belongs under the table: "That is my reading of the supplied explanation, not a measured audience finding." What the table does establish is what the series is for. The family resemblance, in its words, "is deeper than [the palette]. It is customer agency carried through the experience." Four films can look different, sound different and run different lengths, and still make the same promise about what happens to what you say.

The films, one by one

Here they are in the order a viewer would meet them. For each one: its job, how it tells its story, how it connects to the person watching, and what the reviewers made us change. That last part is where most of the learning is.

This is the film about the material, and you watch it work. A glowing bead of tooth-coloured resin fills a chipped edge on a teaching model. Then three beats, one word each on screen. Shape: the instrument carves it back to the planned edge. Colour: chosen with you, from the shade guide. Surface: a matte finish turns glossy under the polisher, and, as the page notes, can be polished again, because it needs ongoing care. The camera pulls back to the model on the consultation table, where fine proportion lines draw across the teeth, and the narrator says: "Shaped around what you'd like to keep." The end card asks the series' question, What would you keep?, while the narrator gives the offer and says it starts with a discovery call.

It's built to play completely with the sound off. Every word on screen names something the viewer may have seen in their own mirror. And it doesn't try to be the whole story. It hands the viewer on to the main film's human question.

It was also the film that failed hardest, and I'd want anyone making AI video to know why. The first cut was, by any ordinary standard, gorgeous: macro stills of resin, shaping, contour, shade tabs and polish, one word per cut in the brand's gold italic. The judge that listens to the soundtrack gave it 8.8 and called it "a masterclass in quiet, premium service advertising." The cold viewer, a fresh model playing a woman with a chipped tooth watching with the sound off, gave it 8 and found nothing misleading. The harsher judge, which reads the keyframes and the transcript, gave it 6, and its first reason was the one that mattered: "the craft shots never connect to what a viewer notices, such as a chipped or short edge."

My thinking partner later put the difference as two sentences a film could be saying. "Look at the sophistication of our craft." Or: "Here is how that craft relates to the thing you care about." In its words, "Both can be visually impressive. Only the second closes the intended connection."

The first repair tied every beat to something a viewer might notice, and it still scored 6. As the build record put it, nothing visibly changed on screen "in a film called 'Up close'." The tools were posed beside finished surfaces, not working. The second repair made the change happen inside the shot: the resin lands and fills the chip, the edge is carved, the surface turns from matte to gloss. It still scored 6, this time for a twelve-second end card sitting static in a film that was running nearly half as long again as planned. The third repair cut it back. Cut 4 passed: 7 from the harsh judge on both shapes, 9.2 and 9.3 from the listening judge, 8 from the cold viewer, no false impressions.

Two things about that sequence are worth holding onto. First, an average would have shipped the first cut. 8.8, 8 and 6 average out to a comfortable pass. The bar was 7 or more from every judge, and that one word is what stopped a beautiful film that didn't fit. Second, look at which reviewer moved. The cold viewer gave 8 in every round, before and after the repairs. The listening judge liked it from the start. All three repairs came from the specific reasons of the one judge that kept saying no. The full story of those three rounds, and why they're a lesson about fit rather than finish, is told in The Taste Harness.

You've seen how it opens: the birthday photo, the pinch-in on her own laugh, the tooth circled in gold. The page calls that "an 'I see you' moment in the first three seconds", and gives the reason: "everyone has a photo they zoom into." She isn't described as self-conscious. Nobody says "do you hate your smile?" She does something the viewer has done, and the viewer recognises it before they know what's being sold.

The gold ring carries the film. It cuts from the photo to the same tooth on a teaching model, where she points: "This one." The dentist starts with two questions, the series' own: What would you change? What would you keep? Her answer ("Could we change the shape of it… but keep the colour close to what I have?") writes itself into the consultation notes as CHANGE and KEEP, and her own shade tab is set aside. That's the being-heard test passing on screen. The colour she wants kept is kept, and nobody reaches for a whiter shade.

Then come the checks a photo can't show (teeth, gums, bite) and a line that's there on purpose: sometimes veneers aren't the first step. If she goes ahead, the smile design starts from her answers. CHANGE and KEEP ride on the design lines, and the plan is made across her whole smile, so that this one tooth sits in balance with the others. The page calls that "the honest bridge from one tooth to a package for 8–10." The narrator says plainly what composite veneers are (tooth-coloured resin, shaped onto your own teeth), then says the offer out loud: two spring packages for eight to ten upper teeth, one shaped by hand and one through a clear mould, with both prices spoken. Then the discovery call.

And then it ends where it began. The same photo, unchanged, pulled back to the whole table of friends. In the page's words: "Nothing about her has changed. The film is about being heard, not about a result." The small ask at the end matches the film's first move: "Tell the team the one you keep noticing."

The page's note on how this film links to the viewer borrows a test from the brief's research: "Where did you see the person's question affect what happened?" Its answer: "Here it happens on screen, twice." Once in the notes, once on the design lines. It's built for the top of the offer page (wide), Facebook and Instagram feed ads (tall), and YouTube.

Two revisions changed three things, and all three are about honesty as much as craft.

• The bridge. The brief I'd supplied had warned against exactly this: don't show someone asking about one small chip and then jump straight to a package for eight to ten teeth, as if the package were the inevitable answer. The early cut didn't explain the jump, and the harsh judge said so. The fix rounds added the reason on screen: the plan is made across the smile so one tooth sits in balance. That became a lesson for every future film, "size the example to the offer": if the example is smaller than what's sold, bridge the gap honestly on screen.
• The price, said aloud. An earlier cut showed the prices on screen for about four and a half seconds and never said them. Anyone listening with their eyes elsewhere never heard an offer at all. Across the series the cold viewer caught the same gap twice, with a line any marketer should find chilling: "I'd ask about pricing on the call." The final cut says the prices.
• What composite is. The narrator now says it, in one plain clause, instead of assuming the viewer knows.

Cut 3 passed with 8 from the harsh judge on both shapes, 9.7 (wide) and 9.2 (tall) from the listening judge, and 9 out of 10 from the cold viewer, with no false impressions. The code that measured the finished cut, which is Code as Proxy's story, clocked it at 65.3 seconds against a 60-second budget, with nothing on screen standing still except the final card.

Two packages at two prices create a problem most offer pages never notice: people assume the dearer one is the better one. This film exists to stop that assumption without arguing with it.

It opens on the question people actually ask, in a customer's voice: "Two packages. What's actually the difference?" On a teaching model, one front tooth is short and chipped. A fine gold line draws its smile design, with two small notes beside it: Colour · keep and Shape · this edge. One smile design, agreed first. Before any method or price appears, the viewer has seen that the plan starts from what the person wants kept.

Then the gold line becomes a seam and splits the screen in two, into the offer page's own package cards:

Reconstructed from the review page's description of the film. The prices are left out here.

Every difference lands in the same row on both sides at the same moment, so, as the page says, "the viewer compares by looking across." And then comes the timing decision that does the most work in the whole series. Just as the two prices appear side by side, the narrator says: "Neither one is better. It depends on your teeth, and what you'd like to keep." That line lands at exactly the moment a viewer would otherwise read the dearer package as the better one. Said five seconds earlier, it would be a platitude. Said on the prices, it answers the thought as the viewer has it.

It closes on the customer's second question, "So which one suits me?", and the answer is that that's what the dentist is there to explain. The same smile design sits under both methods, so the choice reads as the viewer's own, made with the dentist, and not as a sales ladder. The design line that carried "keep" is what split the screen, so even the comparison is built out of the person's preference.

It's for people who already looked at the packages: retargeting visitors to that section of the offer page, sitting beside the two package cards, and answering "what's the difference?" messages. One revision changed two things. It was cut from 43 seconds to 38, and the two methods were rewritten "in matching terms", which sounds like copy-editing until you think about what it protects. If one method is described in richer, warmer language than the other, the words do the upselling the prices were stopped from doing. Cut 2 passed: 9.5 (tall) and 9.4 (wide) from the listening judge, 9 out of 10 from the cold viewer, no false impressions, and 7 from the harsh judge on the one cut its content filter would agree to score.

It's the film a series planned on instinct would leave out, and it may turn out to be the most useful one. It's for the person who watched the main film, nodded, and didn't ask. In the page's words, "It's built for the person who's been putting it off because they don't want to be sold a big treatment."

A man at his kitchen table, seen from behind, fills in the clinic's real enquiry form. He ticks Chipped or worn teeth and Morning. "It's the front ones. They've worn down flat, over the years." Then a single line: asking doesn't commit you to anything.

Then his request travels. It turns up on a call-back card from the appointment team ("Your request: chipped or worn teeth · morning"), and that's where he asks the question everyone has and nobody puts in a form: "And what's it gonna cost me, just to come in?" The narrator answers him straight: "They'll tell you any consultation fee before anything's booked." The rest ticks in: a request isn't a booking, and you can choose not to go ahead. Then he meets the dentist, and his own words are in the notes: "The front ones. Worn down flat." The dentist looks at teeth, gums and bite before anything else, and the options include something to address first.

The quiet device in this one is a path along the top of the frame:

That path is the film's whole argument in one graphic. The steps you take by asking are small and solid. The steps that cost real money are visibly optional, and they only light up when he decides. The end card then says plainly what's on offer this spring: two composite-veneer packages for eight to ten upper teeth, and the price they start from. Seen whole, the film has swapped the usual order. Most ads get you excited about the result and leave the process vague. This one makes the process the thing you can see, and leaves the decision where it belongs.

It widens the series, too. A man in his fifties with worn-down front teeth is a different viewer from the main film's woman with one short tooth, and the series needed him. It runs as retargeting for people who watched the main film but didn't enquire, in follow-up emails, and on the offer page next to the enquiry form.

One detail about that form deserves its own paragraph. When the build agent later went back over every review round, it found that the generated imagery across the series was, in the harsh judge's phrase, "the stock vocabulary of premium cosmetic dentistry": beautiful, and interchangeable with any other clinic's. The distinctive parts were the clinic's own things: the dentist's real portrait, its own lines, and the real enquiry form. I've argued before, in The Website Is the Clue, that a company's enquiry form "is public evidence about how the product currently works." Here it turned out to be the most recognisably theirs frame in the whole series. Generated images carry mood. Distinctiveness comes from the company.

Its one revision made the film say plainly two things it had only implied. The cost question is now answered in the narration, and the packages are named on the end card. Cut 2 passed: 9.4 (wide) and 9.2 (tall) from the listening judge, 9 out of 10 from the cold viewer, no false impressions, and 7 from the harsh judge on the one cut it would score.

Made with code, not a film crew

The review page ends with a short disclosure, and I think it's the right default for anything made this way:

"The films are made with code, not a film crew. The motion design, typography and editing are code; the stills are generated (people shown are examples, teeth are teaching models); the narration uses [local] voices; the music was composed for these films. Each film was built by its own AI builder, then reviewed by three independent judges, one of which listens to the sound, and revised before it goes up here."The series' review page

In practice, "made with code" means each film is a web page with timing attached. The builder writes the scenes as HTML, styling and animation, and HeyGen's open-source HyperFrames renders them frame by frame: "same input, same frames, same output."⁸ That sounds like plumbing, but it's what made the repairs above possible. A film that's code holds still between passes, so a fix round can change the end card and leave everything else exactly as it was. Why that matters well beyond marketing is the argument of Code as Proxy. Here it meant four films could each be rebuilt up to four times in a day without drifting into different films.

Each film had its own builder, an AI agent working in its own folder from a kit: the brand-film skill harvested from films we'd made earlier for other businesses, the offer's facts, my research brief, a set of guardrails, notes from my own books, and the clinic's website harvested into material (its pages, its images, its fonts and colours). The four builds ran side by side and didn't see each other's work. That turned out to matter. Three of the four builders independently found the same device, the person's words travelling through the service, which is a better sign that it's a real idea than any one builder finding it.

The voices came out of the same loop. I'd rejected a synthetic voice on an earlier film that week, and the system took that further than I'd asked:

"When I've given my thoughts and rejections, they've ended up in the prompting, obviously, but they've also ended up in the testing. It tests for voice quality, and it's said, well, this is an Aussie marketplace, let's make sure we've got Aussie voices, and it's gone and tested those voices."Me, talking it through afterwards

A listening judge picked the local voices. Here's the honest part: when the films passed, I hadn't heard those voices myself yet, and the build agent's handover said so plainly, under "Still for you to check". As my thinking partner put it, the records "prevent us from treating a high listening-judge score as equivalent to a confirmed human preference." The principle it came from is sound, and later went into the skill as "cast the voice from the audience's place". But "local Australian voices" is this audience's application of that principle, not the rule. A business whose customers are mostly recent arrivals, or overseas, would cast differently.

Three reviewers, a bar, and a gate

Every cut of every film went to three reviewers, each a different kind of instrument:

• The harsh judge, a Claude Opus model reading keyframes and the transcript, scoring against a written brief.
• The listening judge, a Gemini model that hears the soundtrack.
• The cold viewer, a fresh model given no context and a persona, marked against an answer key.

The judge brief and the answer key were written before any film existed, so the films couldn't bend the target. The answer key included a list I'd recommend to anyone advertising a treatment. A film had an honesty gap if the viewer came away believing that "a specific result or look is promised or typical; the film showed a real patient or their result; veneers suit everyone or all 8–10 teeth must be done; the consultation or discovery call is free; requesting a call books treatment; composite is permanent", and so on down the list of things the film had no right to imply. The bar for every film was 7 or more from every judge, and no honesty gap.

The gate was stricter than the bar. When I asked for the review page, I'd told the agent to put the films up as they finished. Then I stopped the upload:

"No — only final, reviewed videos that are ready."Me, to the build agent

So the page went live with four films marked "Coming soon", and each one appeared only after its fix round and a fresh round of judging. The page explained the work. The review loop approved it.

And the cost, since everyone asks: the build record puts each build or fix round at roughly $2.60 to $8.75 of model time at list prices (it actually ran on a flat subscription). It doesn't total the series, and it doesn't count my attention, which was the scarce input. But the shape is clear enough. The expensive part of this series wasn't rendering or rebuilding. It was knowing what to ask for and when to say no.

The scorecard, read honestly

Here are the four films as they passed:

From the build record and the review page. Harsh-judge scores show wide / tall where both exist; where there's one number, its content filter refused to score the other cut. Every film had no false impressions from the cold viewer.

It would be easy to read that table as "the films scored nine-and-a-half". Don't. Read it by column.

The listening column is the least informative, even though it has the biggest numbers. Across every round of every film, the listening judge scored between 8.5 and 9.7. It gave the scroll-stopper's failed first cut 8.8. The build agent's verdict on it was fair: it "confirms the sound is fine but doesn't separate a weak film from a good one." It earned its place on voices and mix, and nowhere else. A judge that likes everything is telling you about itself.

The harsh column is where the films were made. "Opus's notes drove every real improvement," the build record says: tie each shot to something a viewer notices, show the change instead of posing it, say the price aloud, keep the end card short. Its lower numbers aren't a worse result than the listening judge's higher ones. They're the only numbers in the table that ever said no.

The cold viewer caught what both judges missed, the commercial gap: people finishing a film without knowing what it cost. And its answer key, written before any film existed, is where the honesty gaps were defined, which is what "no false impressions" in every row is measured against.

The gaps are information too. The harsh judge's content filter refused about one call in three on these clinical images. A retry usually worked, but not always, which is why two films carry one harsh-judge number. A refused score isn't a pass, and the table shouldn't pretend it is.

The build agent also counted which criteria the harsh judge had marked under 7 across its seventeen verdicts. Pacing came up 12 times, distinctiveness 10, connection 9, sound 8, and the hook and the demonstration 6 each. That's a useful map of where AI-built marketing film fails first. Every builder ran long (43 seconds against a 35-second target, 67 against 60, 27 against 18). The imagery looked like everyone else's. And the films connected to the viewer less often than they looked like they would. None of those is a rendering problem. All of them are judgment.

Last, the obvious caveat, stated once and meant: every reviewer here is a model. "No false impressions" means a model given a persona found nothing misleading, not that a patient didn't. Why three different judges aren't three votes, and how you learn what each one can and can't perceive, is Code as Proxy's subject. For this series, the practical rule was simple: trust each judge only on what it can actually perceive, and require every one of them to pass.

The checklist ships with the films

The review page didn't stop at the films. Near the end, under the heading "Before posting", it carried a short checklist. That's the part of the deliverable I'd most like other agencies to copy, because it puts the governance in the same package as the creative, written by the same process, before anyone has a chance to post a film without it. Generalised, it said:

• Adults only. Target 18+ on every platform, and mark the videos as not made for kids on YouTube.
• Comments and reviews off on the ads. "Patient comments about results count as testimonials."
• Terms one tap away. Link each post to the offer page, which has the full terms, benefits, limitations and risks.
• Prices with dates. The films show "total package price", the number of teeth and the offer's dates. "If the offer is extended, update the end card rather than calling it limited."
• The practitioner's details. Name, photo and registration line exactly as the offer page shows them.
• Recovery information. The offer page covers risks and care. "Adding a line on what recovery feels like (adjusting to a new shape; when to call) would complete it."

For this article I had that checklist tested, line by line, against the published guidance. Most of it holds up well. The guidelines say advertising for these procedures "must not be targeted or directed at people under the age of 18 years", and on social media must be identified as adult content.¹ Meta separately requires ads for cosmetic procedures to be "targeted to people at least 18 years or older."⁹ On comments, the regulator's own suggestion is exactly the checklist's: practitioners can minimise the risk of testimonials appearing "by disabling reviews, comments or 'tagging' functions on their social media pages."¹ Risk information can sit behind a link when a social post can't hold it. The broader advertising guidelines want an offer's conditions, including its expiry date and total cost, stated or linked, and warn that phrases like "for a limited time only" can be unlawful when tied to claims that someone's health will suffer.¹⁰ "Update the end card rather than calling it limited" is the right instinct, and the practitioner's registration details are required whenever an ad names who performs the procedure.¹

But the checklist was a first draft, and the test found gaps in it. They're worth listing because they're the kind a small business would never catch on its own:

Here's the checklist revised with what the test found. It's a better list, and it's still a draft for a professional to review, not a compliance opinion:

What I take from this is that the checklist belongs in the deliverable, and it needs the same review loop as the films. A film series that arrives with a draft checklist gives the owner something to check and correct. A series that arrives without one hands them a risk they can't see.

Measure enquiries, not views

The review page also said what success will mean, before any film had an audience:

"Measure enquiries, not views. The number that matters is discovery calls from people who already understand what they're asking about. The films set that up: assessment first, prices with their terms, and 'sometimes veneers aren't the first step'."The series' review page

Views would reward the wrong film. The scroll-stopper is likely to win on views: it's short, beautiful and built to be cheap and broad. The first-step film is shown to a small audience, the people who watched and didn't ask, and it's the one closest to a conversation. Rank them by views and you'd put your money in the wrong place. Eric Ries named this trap years ago: most readily available numbers "might make you feel good, but they don't offer clear guidance for what to do."¹⁴ The ad platforms agree when you ask them the right question. Google's conversion tracking starts from the premise that "you choose what you identify as valuable, like a purchase, sign-up, or phone call."¹⁵ For this series the valuable action is a request for a discovery call, so that's the conversion to set up.

There's a regulatory reason not to chase reach here too. The guidelines tell practitioners to "consider the frequency of their advertising and social media posts", because excessive posting may create "the perception that it is normal to have higher risk cosmetic procedures."¹ A campaign measured on views is pushed towards exactly the frequency the guidance warns against.

The sharper part of the page's measure is its qualifier: "people who already understand what they're asking about." That's what the films are for. Someone who has watched the series should arrive knowing there's a consultation before anything else, roughly what it costs, that there are two ways of doing it, and that asking doesn't commit them. One simple way to count it: whoever makes the call-back writes one line per enquiry, answering a single question. Did this person already know what they were asking about? Enquiries alone tell you the films reached people. Informed enquiries tell you the films did their job.

Two cautions keep that honest. First, it's the offer's measure, not the brand's whole scoreboard. The advertising effectiveness research from the IPA is blunt that "short term metrics alone can be misleading", and a spring offer is a short-term thing.¹⁶ Second, the measure was written down before there was anything to measure, and that's deliberate. In the language of Code as Proxy, the intended and delivered accounts of the main film are filled in. The observed account says "not yet measured", and it's defined in advance so that a good-looking number can't quietly redefine success afterwards.

What the making left behind

Once all four films had passed, I asked the build agent the question I think every production job should end on:

"What have we learned from the review process, and the marketing document we wrote? What extra ideas need to go into the original prompting — not just for this sub-project? What can we better reflect in the test harness?"Me, to the build agent, after the last film passed

The answer opened with a sentence I'd like every retrospective to start with: "Your review loop worked: every film improved each round, and the notes that drove the improvement point to a handful of general fixes. Nothing below is applied yet." Then it proposed ten additions to the brand-film skill that every future film on my platform is built from. Each one was written as a principle, a reason and a test, not a rule, because an earlier round of work had taught us that prescriptive rules made the films worse. Here they are, with the moment in this series that taught each one:

Notice what isn't on that list: anything about dentistry. None of the ten is a rule about a clinic. Each one is a statement about fit that should hold for the next business, and each carries its reason, so a future builder can tell when it doesn't apply. Number 6 is the one I'd most want a builder working for any business that can't promise outcomes to have read. It's the title of this article because it's the most transferable thing the job produced.

The same answer tightened the words I'd been using loosely in my own prompts. "I see you" became recognition: a specific detail from the viewer's own life in the first few seconds, shown, not described, and specific enough that someone outside the audience wouldn't recognise it. Its stronger form became being heard, with a test anyone can apply: "Where did the person's words change what happened?" The aha became the moment the viewer works out the mechanism for themselves, and can explain how it works without being told. And demonstrate was sharpened to "a change the viewer watches. A tool posed beside a finished result is a still life."

The tests changed too. The cold viewer got two new questions, graded on how specific the answers are: "Was there a moment that felt like someone like you? What detail?" and "When did you understand how it works? What did you see happen?" The rubric split "connection" into recognition and the aha, and made "demonstrate" require a change on screen. Distinctiveness was rephrased as "hide the logo: name what only this company has." And one fix is a warning to anyone writing test personas: the main film's cold-viewer persona had been described as someone whose front tooth looks shorter, which handed the viewer the answer. The rule now is to describe the buyer, not the film's hook.

I authorised all of it, with a condition: make a fifth film with the new prompting and the new tests, and if either wasn't helping, stop, reset and adjust. What happened next, including the fifth film clearing its judges on the first build and my giving its opening six out of ten anyway, is told in The Taste Harness.

The page also left the clinic something the skill doesn't: a list of films not made yet. Shape or colour?, a 20-second piece in which one person separates the two with a shade tab and a model, "a single complete exchange, not the main film cut down." Why 8–10 teeth?, which explains how many teeth show when you smile, why the number is decided in planning rather than up front, and "why fewer, or none, is always on the table." Looking after them: "composite needs care, can be polished again, and can often be repaired." And Meet the dentist: their own voice and hands, filmed in the clinic and cut into the same look. The page says that one "needs a short shoot, and would lift every other film." I like that the code-made series recommends a camera. Code made the series possible for a business with no crew budget. It doesn't pretend a real person on screen is worth nothing.

Step back and look at what the clinic now has, beyond four films. It has a one-page account of its own customers ("people arrive with one tooth"), six principles for how it talks to them, a relationship its films promise, a checklist, a measure and a roadmap. That's most of a brand guide, written by production rather than by a questionnaire. As my thinking partner put it, "The video becomes an instrument for making the business more explicit."

The finished film is the visible output. The more valuable capability is a system that can help clarify a business while making something useful for it, and retain the method for the next business.

Where the idea actually came from

It would make a better story to say the AI found all this in a clinic's website. That's not what happened, and the truth is more useful.

When my thinking partner read the development records against my account of the day, the records contradicted the neat version. The review page's best ideas look like discoveries of the production process. Most of them were in the brief I supplied before any film existed. I said so plainly at the time:

"The original brief was supplied by me, because I knew it would be absent from the client. I generated it with AI ahead of time, from my background knowledge, what we'd already been doing, and a review of my frameworks and ebooks at that point."Me, on where the brief came from

That brief already held the central question, "What would you change, and what would you keep?" It already separated shape from colour, in a line that became the main film's dialogue almost word for word: "Could we look at the shape, but keep the colour close to what I have?" It already asked for no "we understand your needs" narration, only a viewer watching someone's needs becoming understood. It labelled its audience ideas as hypotheses, warned against jumping from one chip to a package for eight to ten teeth, and steered the films away from testimonials and result reveals because of the advertising guidance. It even quoted my own rule from Demonstrate Over Tell: "A demonstration explains. Evidence substantiates."

What production added was different in kind, and I'd argue just as valuable:

• The device. The person's words travelling, onto a call-back card, into the notes, onto the design lines, found by three builders on their own.
• The compressions. "One tooth", "their words travel", "no results, no reveal": the page's phrases, sharper than the brief's paragraphs.
• The distinctions only failure could teach. Craft displayed against craft connected. Posed against working. A length as a budget. A price shown is not a price said.
• The tests that now check for all of it on the next film.

The attempts didn't originate the central ideas. They sharpened them, bounded them and made them testable. My thinking partner's line for this is the right one: "The provenance makes the achievement clearer, not smaller." What carried into a client with no brand guide wasn't a successful video template. A template carries a previous answer. What carried was my way of thinking about communication, the questions worth asking, and production turned it into answers specific to this business. That's a capability you can plan around. "The AI found the strategy" isn't.

There's a second, quieter lesson in the same records. The brief, the films, the review page and our discussion all agreed with each other, because each was made from the one before. In my thinking partner's words, they "are connected derivatives, not four independent confirmations." Their agreement is not evidence. Only the records could tell us where the idea came from, and only an audience can tell us whether it works. The fuller account of that correction is in The Taste Harness.

What this specimen doesn't prove

I've been careful with the claims all the way through, but it's worth putting the limits in one place, because a specimen is only useful if you know exactly what it's a specimen of.

The second item is the one I'd act on first, and it's not a weakness so much as the next piece of work. The films can't establish that the clinic takes notes in the patient's own words. What they can do is show the clinic a concrete version of that promise and ask whether it's true. As my thinking partner put it, "The actual service constrains what the film may honestly show." And in the other direction: "The film can make a desirable service commitment explicit enough for the business to adopt." A small business that couldn't write down its brand promise can end up, via a film, with its promise written down, corrected by the owner, and true. The method for that conversation, a sheet the owner marks moment by moment, is in The Taste Harness.

And the one experiment I'd run first, once there's an audience, comes straight from the films being code. Hold everything in the main film constant, remove only the beat where her words reappear in the notes and on the design lines, and see whether viewers still say they felt the clinic listened. If they do, the device isn't doing what we think. If they don't, those few seconds are the most valuable in the series. Either answer means something, because we wrote down in advance what that beat was for.

That's what I mean by a specimen. In Specimen, Not Prescription I put the line this way: "It proves the method exists. Discovery determines whether, where and how it applies to you." Four review-ready films, a checklist, a measure and a skill prove a small business with no brand guide can get a coherent, honest film series made in code. They don't prove the films work, and they don't prove the same series is right for anyone else.

The moves worth copying

If you run, or market for, a business that can't promise outcomes (a physio who can't promise recovery, an adviser who can't promise returns, a lawyer who can't promise a verdict, a builder who can't promise council approval), most of this series transfers without the dentistry. Here's what I'd take, each with the test I'd use to check it:

1. Give each film one job, for one moment in the decision. Scrolling, curious, comparing, hesitant.Test: can you name the moment each film is for, and where it will run? If two films have the same answer, you have one film twice.
2. End every film on the same small next step. A request, not a booking.Test: is the ask smaller than the hesitation of the person watching?
3. Start in their world. Open on something the viewer already handles: a photo, a form, a bill, a message.Test: in the first three seconds, would a customer recognise the detail before they knew what was being sold?
4. Let their words travel, and change something.Test: could the service have done exactly the same thing without the person's words? If yes, it's decoration.
5. Turn the constraint into the idea. List what you're not allowed, or not able, to show. Then ask what you can show instead.Test: does the film still work with the forbidden thing removed entirely, not blurred or hinted at?
6. Size the example to the offer.Test: if the story starts smaller than what's sold, is the bridge on screen, and is it honest?
7. Put one thing only you have in every film. Your real form, your real people, your own words.Test: hide the logo. What's still recognisably yours?
8. Say and show the offer.Test: would someone watching muted know the price and the next step? Would someone listening with their eyes elsewhere?
9. Ship the checklist with the films, and have a professional review both.Test: could someone post these safely without calling you first?
10. Decide the measure before the data.Test: what number, arriving next month, would tell you the films had failed?
11. Keep a judge that can say no, and require every judge to pass.Test: has any reviewer ever failed a cut? If not, you don't have a reviewer.

None of those needs code. You could brief a human crew with every one. What code changed is the economics: a small business can now have a whole series, built and rebuilt against a bar, where it used to need a crew and a budget most small practices don't have.

Nothing about her changed

Go back to the photo. At the start of the main film she circles one tooth in gold. At the end the same photo pulls back to the whole table, and nothing about her has changed. What changed in between was where her words went. "Change the shape. Keep the colour." Into the notes, onto the design lines, into the plan. The film never promises what her teeth will look like. It shows that what she said mattered.

That turns out to be the right test for everyone in this story. My thinking partner put it in one line: "A contribution has been understood when it changes the subsequent work, not merely when it has been repeated back." For the woman in the film, it's whether "keep the colour" changes the plan. For me, it was whether my rejections changed the next film, the skill and the tests, and the records show they did. For the clinic, it's the next step: whether their corrections change the films before anyone posts them.

When you can't show the result, show what happens to what people tell you. It's the more honest film, and it may well be the stronger one.

Scott Farrell runs LeverageAI, helping Australian leadership teams turn scattered AI experiments into governed capability that compounds. He builds the systems he writes about. scott@leverageai.com.au · leverageai.com.au