GPT-6 Astra Changed Meta Ads Forever!
2026-10-03 · long tutorial / working session · Adil · recording, 10:23 · watch on YouTube
Topic: Studio work & business (also: Pre-production & story, Prompting & AI assistants, Cost, credits & plans) · mode: Tutorial · tags: fiction, prompts shared, one creator's experience
Adil shows how he makes Meta ads with GPT-6 Astra in ChatGPT, connected to Higgsfield and his custom ad skill, which looks for long-running, wide-reach ads in Meta Ad Library. Product ads follow his five-part PROOF copy framework, and service ads the QUEST framework: a quiz, a case study told as a story or an offer backed by a guarantee. He posts new ads as organic Reels first, A/B-tests the winners one element at a time, and says the whole experiment cost about $100.
What is said
The parts of the video and what is said in each (★ — how much it matters to the video).
00:00:00 — Ten thousand ads analysed and three leads; his last marketing video on this channel
- ★★ Adil says he spent weeks analysing 10,000 ads for products and services with GPT-6 Astra, then tested the patterns with his own ads and got his first three leads. — 00:00:00 · GPT-6 Astra
- ★★ Adil says this is his last marketing video on the Higgsfield channel; new marketing videos will be on Higgsfield Marketing (as heard). — 00:00:00
00:00:29 — The PROOF copy framework and the callout diagram; connecting Higgsfield to ChatGPT with the custom skill
- ★★★ From the 10,000 ads Adil distilled the winning copy structure into a five-part framework he calls PROOF, used in all three e-commerce ad formats, static or video. — 00:00:29
- ★★ Adil's first winning e-commerce ad format is the callout diagram: a clean product shot with thin lines pointing out features, explaining how each helps in everyday life. — 00:00:29
- ★★★ To make a callout diagram, connect Higgsfield to ChatGPT, attach Adil's custom ad skill and add the product photo with a short description; GPT-6 Astra writes research and copy, Higgsfield MCP generates the visuals. — 00:01:15 · Higgsfield AI, ChatGPT, GPT-6 Astra, Higgsfield MCP
00:01:42 — How the skill picks models and mines Meta's ad library for angles; keep visuals simple
- ★★ Adil's ad skill picks a model for each task: GPT Image 2.5 for static ads, Seedance for video and Seedream for characters. — 00:01:42 · GPT Image 2.5, Seedance, Seedream
- ★★ On the GPT-6 Astra side, Adil's ad skill analyses the ads already running in the niche, finds what they have in common and uses those patterns for new creatives. — 00:01:42 · GPT-6 Astra
- ★★★ Meta Ad Library shows an ad's reach and how long it has run; Adil says the longer an ad runs and the more reach it gets, the more likely it converts, and builds his skill on that. — 00:01:42
- ★★ Adil says lead costs stay high because competitors share the same basic understanding of the customer and make similar promises; a deeper understanding of why people buy gives different angles to test. — 00:02:15
- ★★ Adil says scrolling Meta's public ad library shows that the highest-converting ads are usually the simplest, so ad visuals should not be overcomplicated. — 00:02:58 · Meta
00:03:29 — Review card and offer card; adapting for Stories and Reels; turning the three formats into video ads
- ★★ Adil's second e-commerce format is the review card, a real customer quote with a name and a rating; the third is the offer card; all three use PROOF, while services need another formula. — 00:03:29
- ★★ Once the ads are approved, Adil asks GPT-6 Astra to adapt them for Stories or Reels, turning one creative into several formats without redoing the work. — 00:03:29 · GPT-6 Astra
- ★★ For video ads Adil uses the same three formats one level deeper: the product shown working instead of a static callout diagram, a UGC-style video instead of a review card. — 00:03:55 · UGC
00:04:21 — A UGC-style ad for the bag
- ★★ Adil says UGC-style ads do not look like typical ads, so the algorithm reads them as interesting content and shows them to more people. — 00:04:21 · UGC
00:04:30 — Why UGC-style ads get shown more; lead-cost arithmetic; ten minutes instead of a week
- ★ Adil's example for an $80 bag: $10 can bring one lead and no sales, or ten leads and three buyers, $240 in revenue. (hypothetical) — 00:04:46
- ★★ Adil says the whole process with GPT-6 Astra and Higgsfield takes about 10 minutes, not a week. — 00:04:46 · GPT-6 Astra, Higgsfield MCP
00:05:15 — Finding clients in Meta Ad Library; the client brief; a $250 package
- ★★★ To find clients without paid traffic, Adil opens advertisers' ads in Meta Ad Library and clicks through the advertiser card to their official website: these businesses already spend money on ads. — 00:05:15
- ★★ Adil says clients pay for AI ads because the service takes the endless question of which ad to run next off their plate. — 00:05:47
- ★★ Adil sends a lead his free client brief template and feeds their answers straight into GPT-6 Astra. — 00:05:47 · GPT-6 Astra
- ★★ Adil suggests offering a client five AI ad videos at $50 each, a $250 package; the price depends on the work, and more variations can turn into repeat business. — 00:05:47
- ★★★ Adil advises posting new ads as organic Reels first, especially on a small budget, to see which ideas get people asking about the product before paying for traffic. — 00:06:08 · Instagram
- ★★ Adil tested his ad skill by posting a few of the videos as Reels; some got more views than others, and people started reaching out about the bag. — 00:06:08
00:06:17 — Testing the ads as organic Reels; first inquiries about the bag
- ★★ Once stronger ideas are found, Adil makes more versions with one prompt; GPT-6 Astra compares the results with the audience research and suggests new variations to test. — 00:06:40 · GPT-6 Astra
- ★★★ For services Adil uses the QUEST framework, with three formats: a quiz, a case study told as a story, and a direct offer backed by a guarantee. — 00:06:40
00:07:01 — Ads for services: the QUEST framework and niching down
- ★★ Adil says niching down is the one lever that raises a service's price: a general Spanish course might cost $100, a business Spanish program a thousand. — 00:07:09
- ★★ Adil advises picking one customer segment and aiming every ad at it, which alone raises the average check. — 00:07:45
00:07:49 — Storytelling ad made by Astra for a marketing agency
00:08:12 — Checking the agency ad against QUEST
- ★★★ Adil's formula for service ads: hook with a pain point, show the end result, and close with a high-value lead magnet. — 00:08:15
00:08:22 — Quiz ad for an educational course
00:08:35 — The services formula, the lead magnet and the A/B-testing rule
- ★★ Adil says a free lead magnet filters for qualified leads who need the service: a free diagnostic for a doctor, a trial lesson in education, a rebranding consultation for a designer. — 00:08:44
- ★★★ Adil's rule: never rebuild a working ad; A/B-test it by uploading the Ads Manager performance report to GPT-6 Astra for variations of the best creatives, changing one element at a time. — 00:08:44 · A/B testing, GPT-6 Astra
00:09:11 — Three numbers to watch; what all the ads cost against traditional production and UGC creators; links
- ★★ When testing ads on paid traffic, Adil watches three numbers: hook rate, CTR and cost per lead. — 00:09:11
- ★★★ Adil says every static and video ad in the experiment cost about $100 in total, all of it spent on iterations to get the examples right. — 00:09:11 · GPT-6 Astra, Higgsfield MCP
- ★★ Adil says $100 buys one video in traditional production, and a typical UGC creator charges $100 to $200 for a single ad. — 00:09:39 · UGC
- ★★ Adil says he will leave the best-performing templates, his custom ad skill, the client brief, the prompts and the setup instructions in the description, all free. — 00:09:39
Threads in this video
- first on record · How do you write ad copy and an offer that converts? — From the 10,000 ads Adil distilled the winning copy structure into a five-part framework he calls PROOF, used in all three e-commerce ad formats, static or video. (00:00:29)
- refine · How do you write ad copy and an offer that converts? — Adil's formula for service ads: hook with a pain point, show the end result, and close with a high-value lead magnet. (00:08:15)
- refine · How do you make ad creatives at volume without a production team? — To make a callout diagram, connect Higgsfield to ChatGPT, attach Adil's custom ad skill and add the product photo with a short description; GPT-6 Astra writes research and copy, Higgsfield MCP generates the visuals. (00:01:15)
- refine · How do you find clients for AI production work? — To find clients without paid traffic, Adil opens advertisers' ads in Meta Ad Library and clicks through the advertiser card to their official website: these businesses already spend money on ads. (00:05:15)
- refine · What should you charge for AI ads and services? — Adil suggests offering a client five AI ad videos at $50 each, a $250 package; the price depends on the work, and more variations can turn into repeat business. (00:05:47)
- first on record · How do you find out which ad creative works best? — Adil advises posting new ads as organic Reels first, especially on a small budget, to see which ideas get people asking about the product before paying for traffic. (00:06:08)
- refine · How do you find out which ad creative works best? — Adil's rule: never rebuild a working ad; A/B-test it by uploading the Ads Manager performance report to GPT-6 Astra for variations of the best creatives, changing one element at a time. (00:08:44)
- refine · What does an AI clip, shot or film cost? — Adil says every static and video ad in the experiment cost about $100 in total, all of it spent on iterations to get the examples right. (00:09:11)
- promised · Adil: put the best-performing ad templates, his ad skill, client brief, prompts and setup instructions in the description (2026) — Adil says he will leave the best-performing templates, his custom ad skill, the client brief, the prompts and the setup instructions in the description, all free. (00:09:39)
- refine · How do you change the aspect ratio of an existing image or video? — Once the ads are approved, Adil asks GPT-6 Astra to adapt them for Stories or Reels, turning one creative into several formats without redoing the work. (00:03:29)
Where the stories that moved here went next
- Adil: put the best-performing ad templates, his ad skill, client brief, prompts and setup instructions in the description (2026) — open (no report in 2026); last: promised, 2026-10-03
Mentioned
Central: Adil's Meta ads skill (GPT-6 Astra) · GPT-6 Astra · Higgsfield MCP · Meta Ad Library · PROOF framework · QUEST framework
Also: Organizations: Higgsfield Marketing
Concepts: A/B testing · lead magnet · UGC
Products: Ads Manager · ChatGPT · GPT Image 2.5 · Seedance · Seedream
Only in the description or credits: Instagram · Meta
← Previous: Editorials · Next → Sora 2 API Shutdown: Best Alternatives and How to Migrate (2026)