Five formats, ranked, with the reasons already written.
Both ad banks are written as talking head, which is format one and the best of them. This page is the other four. The rule underneath it is the one that makes the banks reusable: the reason to buy does not change, only the vehicle carrying it does. A reason that works as a 45-second talking head works as a text-on-b-roll reel, as a single static, and as a six-card carousel, and running the same reason across all four is not repetition, it is coverage. What you must never do is invent a new reason to fit a format. Pick the reason from the bank, then build it here.
Ranked best to worstReasons come from the two banks55 reasons availableNo new claims introduced here
The ranking, and what it actually means for the budget.
1. Talking head, 30 to 60 seconds. Built. 35 reasons on the call bank, 20 on the webinar bank, 34 of them scripted at both lengths.
2. Static reels, b-roll plus text. The same reason with no face. Cheap, fast, and the only format you can ship without booking a film day.
3. Static images. Best for the two-day event, because a fixed date, a time and a price are read faster than they are heard.
4. Carousel ads. Feature-based, tweet threads. Buys the reader's own pace, which suits an argument that needs three steps.
5. Testimonial carousels. Last on the list, and last for a reason: they need approved quotes and they persuade people who are already warm. Retargeting, not cold.
Read the ranking as a production order, not a quality verdict. Format 1 is where the money should go and it needs a film day. Formats 2, 3 and 4 can be built this week from what is already written, which makes them what you run while the film day is being scheduled. Format 5 waits on real quotes and should never be the first thing a cold audience sees.
Format 1 · talking head, 30 to 60 seconds
Built. Do not rebuild it here.
Every scripted reason on both banks is a talking head with a 30 and a 60 second cut, hook to CTA, one reason per ad. There is nothing to add on this page except the three things that separate a talking head that works from one that does not, all three from Jeremy's originals.
What decides it
SOP-26, 27, 35
The three rules
1. The first five seconds carry the payoff or the frustration, never a name, never "hey guys", never "I'm here in my office" (SOP-27). If the reason cannot be said in under ten words, it is not ready to film.
2. One reason, filmed whole, hook to CTA (SOP-31). Post-Andromeda, swapping hooks onto a shared body "lasts for a much shorter duration". Fifteen to twenty genuinely different reasons a month is the floor.
3. Say everything on the list (SOP-35). Vision and ROI, no audacious claims, no hype, no red on screen. The bigger the claim, the lower the quality of the person it attracts.
The text carries the argument and the b-roll only has to be neutral and unbranded. This is the format to run first, because it needs nothing that does not already exist. The discipline is reading speed: one idea per card, maximum nine words, minimum 1.4 seconds on screen per card and 2.2 seconds for any card carrying a number. Sound off is the default, so the text must survive with the audio muted, and the last card must hold for three full seconds because that is the one carrying the CTA.
Spec, once, for all of them
Production
Build spec
9:16, 1080 x 1920. Safe area: keep all text between 250px from the top and 420px from the bottom, because Reels and Stories chrome eats both ends.
Type: Archivo, 800 weight for the card line, 400 for any source line. White on near-black (#0d0f14). One accent only, Picton blue #3DB7E9, used on the single word that carries the turn.
No red anywhere (SOP-35). No stock-footage people pointing at laptops. No music with a drop.
B-roll: screen recordings of a real AI answer, a real SERP, a cursor scrolling a real report. Filming your own screen is both cheaper and more credible than any stock library.
Captions burned in, not auto-generated, because the text is the ad.
Length: 12 to 22 seconds. If a reason needs more than 22 seconds as text, it is a talking head, not a reel.
R1 · Ask the machine about your own category
Call funnelFrom reason 35
Reason: the demo does the arguing. Nobody needs to be told they are missing from the answer if they watch it happen.
Card sequence, 18 seconds
0-2.5s
Ask ChatGPT who the best in your category is.
B-roll: a real prompt being typed. No face.
2.5-8s
Watch it name three businesses.
The answer streams in. Let it actually stream, do not speed it up. Three competitor names visible, blurred only if legally required.
8-11s
You are not one of them.
Hold. Nothing moves. This is the whole ad.
11-15s
That answer is decided by six signals on your site.
15-18s
Free 30-minute audit. You keep the plan.
CTA card, hold 3s. sh.studiohawk.com. Small line: for businesses doing $1M+.
Reason: the exercise does the qualifying inside the creative. Works better as a reel than a talking head, because the viewer does it on the same device.
Card sequence, 17 seconds
0-3s
Last month's ad spend. Times twelve.
B-roll: a phone calculator, real thumb, real numbers.
3-6s
Do it now. I will wait.
Hold on the calculator. Let the pause be uncomfortable.
6-10s
That is your annual rent.
10-14s
It stops the day you stop paying. Someone else sets the price.
14-17s
We will price the channel you own against it. Free.
R7 · Your rankings look fine. That is the problem.
Call funnelFrom reason 30
Reason: the person who is losing has no symptom to look at. The reel shows the green report and the empty AI answer side by side, which no talking head can do as fast.
Card sequence, 15 seconds
0-3s
Your ranking report is green.
B-roll: a real report, all green arrows.
3-7s
Your traffic is flat anyway.
7-11s
Because the answer sits above the ranking.
Split screen: green report left, AI answer right with three names, none of them theirs.
Eight is a month of reels at Jeremy's two-a-week floor for a secondary format. Every one is a reason that already exists in a bank, so if a reel wins, its talking-head version is already written and can be filmed on the next film day rather than commissioned from scratch. That is the point of keeping the reason and changing only the vehicle.
Format 3 · static images
Best for the two-day event, and here is why.
A date, a time and a price are read faster than they are heard. The webinar has all three and they never change inside a cohort, which makes a static the most efficient asset this funnel owns: one file carries the entire offer, and the monthly refresh is two numbers. The rule for statics is harder than for video, because there is no second chance: one claim, one number at most, one CTA, and the whole thing legible at 320px wide in a feed on a phone at arm's length. If you have to zoom to read it, it has failed.
Spec, once, for all of them
Production
Build spec
Sizes: 1080 x 1350 (4:5) as the primary, 1080 x 1080 (1:1) as the fallback, 1080 x 1920 (9:16) for Stories. Build 4:5 first, crop down.
Text: under 20 words total on the image. Headline at 64pt minimum on a 1080 canvas.
Type: Archivo 800 for the claim, 400 for the source line and the date bar.
Colour: near-black ground #0d0f14, white type, one accent. Picton #3DB7E9 for the call funnel, Yellow #FECF40 for the event date bar so the two funnels are visually separable in the ad account at a glance. No red (SOP-35).
The date bar: a fixed strip along the bottom on every event static reading "Sep 23 + 24 · 12:00 PM ET · $7 · live only". This strip is the monthly refresh. Nothing else on the file changes.
Source lines: any static carrying a statistic carries its source in 400 weight beneath it, always. This is the whole credibility difference between these and a generic agency ad.
No stock photography of people. Screens, charts and type only.
Event statics · the eight that carry the offer
Webinar4:5 primary
S1 · the flat statement
HEADLINE: There is no page two of an AI answer.
SUB: You are named, or you are not there.
DATE BAR: Sep 23 + 24 · 12:00 PM ET · $7 · live only
Layout: type only, centred, nothing else. The most reusable file in the set.
S2 · the number
HEADLINE: $40.74 → $70.11
SUB: Cost per lead on Google Ads, 2022 to 2025. Up 72%.
SOURCE: WordStream / LocaliQ
DATE BAR: as above
Layout: the two figures at 140pt with an arrow between them, everything else small.
S3 · the screenshot
HEADLINE: Three businesses get named. None of them are you.
IMAGE: a real, unedited AI answer for a real category, three names visible.
DATE BAR: as above
Layout: screenshot occupies the top two thirds, headline beneath it. The proof is the image, so do not crop it into unreadability.
S4 · the analogy
HEADLINE: You were sold exercises. Nobody gave you a programme.
SUB: Blogs, meta titles, a monthly report. Building toward what?
DATE BAR: as above
S5 · the shortlist
HEADLINE: 95% of winners were on the day-one shortlist.
SUB: A machine writes the shortlist now. Two days on getting onto it.
DATE BAR: as above
S6 · the calendar
HEADLINE: One cohort a month. No replay.
SUB: SEO in an AI World. Two live days with Harry Sanders.
DATE BAR: Sep 23 + 24 · 12:00 to 1:30 PM ET · $7
Layout: the date is the hero here, set at 120pt. This is the scarcity asset and the only one where the date is the headline.
S7 · the price
HEADLINE: $7.
SUB: Two days. Live. The whole framework, not a pitch.
DATE BAR: as above
Layout: the price at 200pt, alone in the top half. The cheapest qualification in the funnel, used as the hook.
S8 · the rented / borrowed / owned card
HEADLINE: Rented. Borrowed. Owned.
SUB: Ads stop when you stop. Social is on loan. Only one of the three is yours.
DATE BAR: as above
Layout: three stacked rows, the third in the accent colour. Reads in under two seconds, which is the whole test.
Eight files, one date bar, and the monthly refresh is that strip. Run S1, S3 and S6 as the opening three: a flat statement, a proof screenshot and the scarcity. S7 is the one to test against S1 for cold, because a $7 price tag as the hook either strips out tyre-kickers or strips out serious buyers, and the answer is not predictable from here.
Audit statics · four for the call funnel
Call funnel4:5 primary
S9 · the study
HEADLINE: Tools beat blogs by 7.5x in AI search.
SUB: We classified 7.5 million pages against 1,203,748 AI referrals. Comparison pages came fifth.
SOURCE: StudioHawk, Format Leverage, August 2026
CTA STRIP: Free 30-minute AI Search audit · $1M+ businesses
Layout: ranked bar chart, five bars, tools highlighted in accent. The only static in the set that is a chart, and it earns it.
S10 · the contradiction
HEADLINE: Statistics get cited 30 to 40% more. Statistics pages came last.
SUB: A statistic is an ingredient, not a format.
SOURCE: GEO, KDD 2024 · StudioHawk Format Leverage, 2026
CTA STRIP: as above
Layout: two lines, the second in the accent. This is the most scroll-stopping line in either bank because it looks wrong until you think about it.
S11 · the ledger
HEADLINE: Last month's ad spend x 12. Circle it.
SUB: That is your annual rent. We will price the channel you own against it.
CTA STRIP: as above
S12 · the green report
HEADLINE: Your rankings look fine. That is the problem.
SUB: The answer sits above the ranking. Your report cannot see that layer.
CTA STRIP: as above
Layout: a real green ranking report, desaturated, with a single AI answer card overlaid on top of it.
Format 4 · carousel ads
For the argument that needs three steps.
A carousel buys the reader's own pace, which is worth having when the reason is a chain rather than a claim. The cost of that is card one: nobody swipes unless card one earns it, so card one is a hook and never a title slide. Rules: six to eight cards, one idea per card, under 25 words per card, the last card is always the CTA, and card two must pay off card one immediately rather than making them wait. Build every card as a standalone 1080 x 1080, because Meta will sometimes serve them individually.
C1 · feature-based: what the six signals actually are
Call funnel7 cards
Reason: the audit is abstract until you list what is being audited. This is the carousel that converts "sounds vague" into "I want to know how we score".
Card by card
1
Six things decide whether an AI recommends your business. Most sites fail four.
Hook card. No logo, no title. The number four is the accent.
2
1. Can the machine actually read your pages, or is your content locked inside JavaScript?
3
2. Does anything on your site do a job an article cannot? Tools earn 7.5x their share of AI traffic.
Source line, small: StudioHawk Format Leverage, 2026.
4
3. Is there evidence on the page, or only claims? Statistics and citations get cited 30 to 40% more.
5
4. Does the rest of the internet agree you exist? The machine checks who else says so.
6
5. Is your category described in the words buyers actually ask in, or the words you use internally?
7
6. Is any of it maintained? A stale answer is worse than no answer. Free 30-minute audit, we grade all six.
CTA card. sh.studiohawk.com. Small line: for businesses doing $1M+. You keep the plan either way.
Primary text
Six signals decide whether an AI names your business when a buyer asks. Most sites we audit fail four of them, and the failures are rarely the ones people expect. Swipe for the list, then get yours graded free in 30 minutes with a senior strategist. You keep the plan whether you work with us or not. For businesses doing $1M+.
Reason: the study reads naturally as a thread because it has a finding, a twist and a conclusion. Set each card as a tweet-styled block on a plain ground. The format signals "someone is telling you something" rather than "someone is selling you something", which is exactly why it works and exactly why it must not carry a hard pitch until card six.
Card by card
1
We got sick of guessing what AI search rewards, so we measured it. 1,203,748 AI referral sessions. 600+ businesses. 7.5 million pages. Here is what actually gets cited.
Tweet-card styling: avatar, handle, plain text, no imagery. Real StudioHawk handle, never a fabricated account.
2
We classified every page by format, then asked which formats earn more AI traffic than their share of the internet says they should.
3
Winner, and it was not close: tools, templates and calculators. 7.5x their share of pages.
4
Comparison pages, the format every agency sells as the AI-search play, came fifth.
5
Statistics round-ups came last. Which looks like a contradiction, because pages carrying statistics get cited 30 to 40% more often. It is not. A statistic is an ingredient, not a format. A page whose only job is holding numbers is a summary, and summarising is the one job the machine never needed you for.
The longest card in the set. It earns the length; do not trim it.
6
Full study is public. If you want to know where your own content sits on that list, we will grade it free in 30 minutes.
CTA card. sh.studiohawk.com.
Primary text
We measured 1.2 million AI referrals across 600+ businesses to find out which content formats AI search actually cites. Tools and calculators won by 7.5x. Comparison pages came fifth. Statistics round-ups came last, and that last one only looks like a contradiction. Swipe. For businesses doing $1M+.
Only ever build a tweet-thread card from a post that was genuinely published, or from plain text on a plain ground with no platform styling at all. A fabricated screenshot of a tweet that was never posted is a real exposure, and it is the kind of shortcut that costs an ad account. If the thread has not been posted, post it first, then screenshot it.
C3 · feature-based: what two days actually covers
Webinar8 cards
Reason: the single biggest objection to a paid event is not the $7, it is the two days. This carousel answers "what do I actually get for that time" with an agenda, which no video does as well because the reader wants to scan it.
Card by card
1
Two days is a lot to ask for $7. Here is exactly what is in them.
Hook card. Names the objection instead of dodging it, per SOP-27.
2
Day one: why your rankings can look fine while the traffic goes flat, and what replaced them.
3
The citation map: which businesses AI names in your category, and the six signals that decide it.
4
The rent ledger: what you are paying a year for traffic that stops the day you stop.
5
Day two: the build. Which formats to make, in what order, with the study behind the order.
6
How to brief it, whether you have an in-house team, an agency, or neither.
7
Live Q and A. Not a webinar with the questions turned off.
8
$7. Sep 23 and 24, 12:00 to 1:30 PM ET. One cohort a month, live only, no replay.
CTA card. webinar.studiohawk.com. The date is the accent.
Primary text
Two days is a lot to ask, even for $7. So here is the agenda, card by card, before you decide. Day one is what changed and how to see it in your own category. Day two is what to build and in what order. Live Q and A, one cohort a month, no replay. Sep 23 and 24, 12:00 PM ET.
Last on the list, and it needs something the other four do not.
Every other format on this page can be built from what is already written. This one cannot, because it needs approved quotes from named clients, and those have to come from a real source with permission attached. What follows is the structure and the shot list, with the client results that are already public on the hub filled in and the quote lines left explicitly blank. Do not write a quote to fill a gap. A fabricated testimonial is not a copywriting shortcut, it is a legal problem and an FTC one, and it is the single fastest way to lose an ad account. Pull the quotes from the review wall on the VSL page, from the results published on the confirmation page, or ask the account leads and get it in writing.
T1 · results carousel, the numbers that are already public
Retargeting only6 cards
Reason: proof stacked in one place, for people who have already seen a cold ad and are deciding whether this is real. Cold audiences do not care about your clients yet, which is why this format ranks fifth and why it should be excluded from cold sets entirely.
Card by card
1
We ran the audit on ourselves first. Then on these.
Hook card. The self-audit is what makes the rest of the cards land as evidence rather than boasting.
2
Case-Mate: 95 percent.
Set the metric exactly as it is stated on the call bank, reason 22. Do not restate it more impressively than the source does.
3
First Fence of Georgia: top-three rankings up 172 percent.
A local service business, which is the objection-handling card. Label it as such on the card: "not a big-brand game".
4
"Great to work with! Five stars, no notes." Ryan Dobrin, Content Marketing Manager, Sylvane.
CLEARED. Published on Sylvane's own case study page, so no fresh approval needed. A second cleared quote exists: Ashlie McKinnon, General Manager, Woodbury Furniture, from the SOURCE manuscript. Those two are the whole approved set; anything else still needs sourcing.
5
The same six-signal framework runs for QuickBooks and Weber as for a fence company in Georgia.
Only name clients whose relationship is already public. Check before it ships.
6
Free 30-minute audit, graded on the same six signals. You keep the plan.
CTA card. sh.studiohawk.com. For businesses doing $1M+.
Audience
Retargeting only. Exclude from all cold sets. Suggested audience: 3-second video viewers of any format 1 or 2 asset, plus site visitors, 30 days, excluding anyone who has already booked. This is the same exclusion logic already in use on the Hammer Them library.
Reason: the strongest thing a paid event can show is what someone said after attending the last one. This is genuinely blocked until a cohort has run and been surveyed, and it should stay blocked rather than be filled with client testimonials that are about the agency and not the event.
Structure, for when the quotes exist
Card 1: hook, "What people said after the last two days." No claim on this card.
Cards 2 to 4: one attendee quote each, with a first name, a role and a company size band. Quote only, no design flourish, set large on a plain ground.
Card 5: CTA. $7, next cohort's date, live only, no replay.
Collection method: the two-question survey already proposed for the confirmation page is the right instrument. Ask it immediately after day two while it is fresh, ask permission to use the answer in the same form, and store the permission with the quote.
Status: blocked until the September 23 to 24 cohort has run and been surveyed. That makes it an October asset at the earliest. Nothing else on this page depends on it, so it should not hold up the other four formats.
How to run it
Which format goes where.
Cold, warm and the split
Media plan
Cold
Formats 1, 2 and 3 only. Talking head carries the budget, static reels fill the gap while a film day is being scheduled, statics carry the event because the date and price are the offer. Feature carousels (C1, C3) can run cold and often do well, because they answer "what is this" before asking for anything. Testimonials never run cold.
Retargeting
Formats 4 and 5, plus the Hammer Them library. This is where testimonial carousels earn their place: the viewer already knows what the offer is and is deciding whether it is real. Exclude anyone who has converted.
The split, if you are starting from nothing
60% format 1 (talking head), 20% format 2 (static reels), 15% format 3 (statics), 5% format 4 (carousels) on cold. Everything in format 5 sits in retargeting on a separate small budget. Revisit after the first 30 days on actual cost per booked call, not on cost per lead, because the whole point of the $1M+ bar and the $7 price is that they buy a worse lead number and a better call.
Naming and tagging
Every asset above carries a tag parameter and a file name. Keep both. The tag is what makes the dashboards able to answer "which reason worked", and the file name is what stops the ad account becoming unreadable at 40 assets. Format prefix, number, name, length: SH_Reel_R2_SameLandlord_16s.
Verification flags
What is not settled.
Three things on this page need a human before they ship.
1. Testimonial quotes do not exist yet. Two quotes are now cleared and named: Ryan Dobrin at Sylvane, published on their case study, and Ashlie McKinnon at Woodbury Furniture, from the manuscript. Everything beyond those two still needs sourcing. Fill from an approved source or cut the card. This is flagged again on the cards themselves so it cannot be missed by someone building from the copy alone.
2. Client names need a public-relationship check. Case-Mate, First Fence of Georgia, QuickBooks and Weber are all already named on the call bank, so they are as verified as anything else on this hub, but "already on the hub" is not the same as "cleared for paid media". Confirm before a name goes into a paid ad.
3. The tweet-thread cards must come from real posts. C2 is written to be posted first and screenshotted second. If that ordering is inconvenient, run the cards as plain text on a plain ground with no platform styling, which loses a little of the effect and none of the argument.
Everything else here is either a production spec or a reason that already carries its source on the bank it came from. No new claims were introduced on this page.