By Green Flag Digital, from Joe Robison's notes and recordings. Updated November 15, 2026.
When a developer asks an AI assistant which agent framework or eval tool to use, the assistant splits the question into smaller searches, pulls passages from what comes back, ranks them and writes from the few that survive, usually as a list. The goal for an AI company is to be one of those passages. Our AEO for AI companies page is the short version.
Buyers already shop this way. In GFD's own recorded sales calls, four buyers told us they found us by asking an AI assistant. One said: "I went to AI, and I researched what's digital PR, and suggest me an agency, and your agency came up." Another said they used every AI tool and asked about 90 different questions, and we came through in the top ten. A third asked ChatGPT to "list some good agencies in the U.S."
How does ChatGPT decide which tools to recommend?
It searches, reads and writes from the best passages. Google says AI Overviews and AI Mode may use "query fan-out", several related searches across subtopics. OpenAI says sites that block its search crawler, OAI-SearchBot, will not be shown in ChatGPT search answers.
In a May 2026 training recording, Joe explained latent space to his team: Google and the LLMs map language so that keywords and topic coverage both matter, but computationally it is about how information is plotted and connected. Google Ads, Meta Ads, SEO and LLM visibility are all run by Big Tech at massive scale.
For a new tool, a name that appears on few pages, next to few of the words buyers use, sits far from the question.
Why do AI answers come back as lists?
Because a list is the most useful short answer to "which one should I use?". Rehearsing a talk in 2024, Joe argued that modern AI mostly answers in lists, that lists bridge broad thinking and action, and that a person's own list, in their zone of genius, is something AI can't match. Lists make money too, power rankings and composite indexes among them.
The engines often build their list from lists people already wrote: roundups, comparison pages and review sites. A tool absent from those starts behind.
Who writes the lists the engines read?
Often, the companies on them. When we read the sites of fifteen SEO, content and PR agencies in November 2026, at least five ranked themselves first on their own "best agency" lists without saying so on the page. One described inclusion in list articles as the largest factor in AI recommendations.
A list with a stated method, a "best for" on every entry and its owner's place disclosed is the kind worth earning.
When Joe tested a "best agencies" prompt for one B2B category in December 2025, the first answer leaned on a list where the publisher ranked itself first. GFD's own roundup, which included GFD and said who each firm suits, was cited next and several times more, while it sat on page three of Google.
Does Bing matter for ChatGPT?
In that same test, the roundup ranked first in Bing while it sat on page three in Google, and ChatGPT's picks followed Bing. A Bing visibility check should be a standard line item in an AEO audit, because it is a cheap way to explain ChatGPT citations. It is one test, so check your own category.
Why do fast-moving AI categories lose answers so quickly?
Pages go stale fast, as Joe told his team in a December 2025 training session:
Whenever we do fresher topics like AI model ranking, if we don't update that page, if it's a trending keyword, it's probably going to drop. It's a very new, evolving keyword, so we probably want to show the decay on the clicks over time and build a refresh cycle.
AI tools ship weekly, so stale pricing or model support is the first thing to fix.
What should an AI company change first?
- Let the crawlers in: OAI-SearchBot and Google's crawlers, in robots.txt and at the CDN.
- Serve docs, pricing and integrations as HTML a crawler can read without JavaScript.
- One page per buyer question: an integration page per model, cloud and IDE; a fair comparison per real rival.
- Earn places on the lists the engines cite, never a paid rank.
- Refresh on a schedule, dating every page.
- Measure on a fixed prompt set, across runs and engines.
Our AEO work for AI companies starts with the first four.
Where is this heading?
In a July 2026 voice memo, Joe argued that AI, search and social systems are all moving toward a world model that tries to validate reality, and that brand work is balancing an accurate, positive representation across all three. In his words: "The biggest, loudest, and most controversial systems of today point to the reality of tomorrow."
FAQ
Can you pay to be recommended by ChatGPT?
Not in the answer itself. Labeled ads are fine where sold; a hidden paid slot in a ranked list raises disclosure questions under the FTC's Endorsement Guides, 16 CFR Part 255.
Do llms.txt files get you recommended?
Little evidence says so, and Google says its AI features need no special markup.
How long does it take to show up in AI answers?
It depends on each engine's refresh cycle and on how fast outside sources mention you, so track a fixed prompt set weekly and judge the trend.
Data
| Metric (unit) | Value | Source | Last verified |
|---|---|---|---|
| Searches an AI Overview or AI Mode answer may issue (count) | several, "query fan-out" | Google Search Central | 2026-11-15 |
| Sites shown in ChatGPT search answers if they block OAI-SearchBot | none | OpenAI crawler docs | 2026-11-15 |
| Agencies of 15 ranking themselves first on their own lists without disclosure (count) | at least 5 | GFD teardown | 2026-11-15 |
| GFD buyers who said they found GFD by asking AI (count) | 4 | GFD sales calls | 2026-11-08 |
| Google results page of the GFD roundup, incognito | 3 | Joe Robison, recorded walkthrough, Dec 2025 | 2025-12-02 |
| Bing rank of the GFD roundup, incognito | 1 | Joe Robison, recorded walkthrough, Dec 2025 | 2025-12-02 |
Methodology
This explainer was written on 2026-11-15 from Joe Robison's recorded training sessions (May 2026, December 2025), voice memos (2024, July 2026), his December 2025 ChatGPT test, and GFD's November 2026 teardown of fifteen agencies. Recorded passages were cleaned by cuts only: fillers, false starts and screen narration. Words are counted with `marketing/site/carbon.mjs` above the closing marker.
References
- Google Search Central, AI features and your website (query fan-out). https://developers.google.com/search/docs/appearance/ai-features
- OpenAI, Overview of OpenAI crawlers (OAI-SearchBot). https://developers.openai.com/api/docs/bots
- Electronic Code of Federal Regulations, 16 CFR Part 255. https://www.ecfr.gov/current/title-16/chapter-I/subchapter-B/part-255
- Green Flag Digital, competitor teardown of fifteen agencies, 2026-11-15.
- Joe Robison, training recordings and voice memos, 2024 to 2026 (GFD Knowledge Index).
History of this topic
- 2023: Researchers coin "generative engine optimization" in a paper on how AI answers pick sources.
- 2025: Joe Robison finds ChatGPT citing GFD's roundup from page three of Google, tracking Bing instead (reference 5).
- 2026: Google's documentation describes query fan-out for AI Overviews and AI Mode (reference 1).