When your prospects ask ChatGPT or Gemini one thing, the mannequin quietly fires a set of conventional internet searches in the background, retrieves the rating pages, and synthesizes the answer from those. The websites that rank for these hidden queries get cited. Those that don’t, don’t. QueryFan generates persona-specific prompts, runs them by each fashions, and captures the precise searches every one triggered. That checklist is your actual AI visibility goal. It’s free.
Key phrases Lists Are Helpful, They Simply Miss Half The Image
Let me be exact about that before anybody writes a livid reply.
I’m utilizing the time period “key phrases” to refer to the “one-shot” queries that go into conventional search engines like google. Sure, I do know we’ve been in a “semantic” world for over a decade, however let’s simply agree on terminology that everybody can observe for now.
The first problem of “key phrase lists” in context to AI search is threefold:
- Usually, queries (prompts) that go into LLMs have a tendency to be longer, multifaceted, and conversational in nature. Conventional searches have a tendency to be extra slender in scope.
- Conventional search is “one-shot.” You do your search, get your information, then do one other impartial search. Queries/prompts on LLMs have a tendency to be conversational in nature and carry the context of earlier tokens.
- The mechanisms that LLMs use for internet search additionally carry personalization context. If the consumer has beforehand acknowledged they are a vegan, they usually ask the LLM about [running shoes], it is extremely seemingly the LLM will carry out a search to accommodate this.
In essence, AI search has turn into a form of “common intent decoder” for customers. These large, multifacted conversations with the AI get damaged down into subsets of solvable queries, which are run in the background as “conventional” searches on Google or Bing, with the ensuing websites used to generate a response. The method is referred to as “Retrieval Augmented Technology” (RAG).
The optimization goal has moved. You are now not optimizing purely for what the human sorts right into a chat field. You are optimizing for what the AI agent quietly searches for on their behalf, in the background, with out the consumer realizing it occurred.
These background queries are what QueryFan captures. They are usually fairly totally different from what the consumer truly requested. They usually are the precise checklist of belongings you want to rank for to seem in AI-generated solutions.
Exhibit A: Reddit Fell Off A Cliff On A Tuesday
The scope and depth of this secret relationship turned clear when Reddit was enjoying meteoric visibility increases in Google, and tragedy struck on September tenth, 2026. In accordance to quotation monitoring information from PromptWatch, Reddit’s citation rate in ChatGPT responses collapsed virtually in a single day. It had been operating as excessive as 15% of all citations. Inside days, it was sitting beneath 2%.
The trigger was unglamorous: Google quietly removed the means to request 100 search outcomes concurrently (the num=100 parameter) from its search API on that date.
Take into consideration what this tells you. Reddit’s visibility in ChatGPT responses tracked Google’s bulk search capabilities, not something Reddit did, not a coaching information replace, not an alignment tweak. The implication is about as delicate as a dropped piano: ChatGPT was bulk-pulling Google search outcomes, Reddit dominated these outcomes at the time, and when the bulk-pull disappeared, so did Reddit’s citations.
AI search surfaces are, in large part, wrappers around traditional search. The “AI” bit is actual (the synthesis, the personalisation, the conversational coherence) however the information retrieval step is remarkably acquainted. Google indexes and ranks the internet; the AI consults that index. Your content material nonetheless wants to rank.
How QueryFan Works
Step 1: Your ‘Conventional’ Key phrases
Your conventional key phrase checklist for the time period “trainers” could incorporate varied urged variations of this time period, from a supply like Google Recommend.
For QueryFan, we are able to merely take the subject of “trainers” and use this as our first step, as we are going to generate prompts round this.
Step 2: Outline Personas
Your personas are how we are going to customise the prompts we generate. This will alter our traversal of the token area, aligning us with coaching information from the tens of millions of communities, discussion board posts, Reddit threads, and web discourse the place actual customers ask actual questions with these identities.
QueryFan sends your persona + subject mixture to the LLM to generate the sorts of questions that persona would truly ask an AI software. Not key phrases. Questions. Actual, conversational, context-laden questions. For the [middle-aged vegan man who just started running] instance, it can produce issues like:
- “Which vegan trainers are good for middle-aged males simply beginning to run?”
- “The place can I purchase vegan trainers on-line in the UK?”
- “What ought to I search for when selecting my first pair of trainers as a newbie?”
Step 3: LLM Choice And AlsoAsked Enrichment
AI conversations department. Somebody who asks about vegan trainers will ask follow-up questions: about value, about manufacturers, about damage prevention. QueryFan passes the generated prompts by the AlsoAsked API to seize the nearest-intent follow-up questions round every one. Individuals Additionally Ask information is the proper instrument right here as a result of it was constructed to mannequin query proximity, which is exactly what you want once you’re attempting to predict the place a dialog goes subsequent.
As an illustration, a search in the UK for “trainers” would floor observe up questions on particular manufacturers, asking how to choose a shoe, and even widespread medical queries.
You can too choose if you want to use ChatGPT, Gemini, or each. Every LLM handles and fan out queries barely in a different way, so if you happen to’re optimising for a selected platform it is finest to get the information from there.
Step 4: Question Fan-Out
QueryFan sends the enriched immediate checklist to GPT-5 with internet search enabled (by way of the OpenAI Responses API) and to Gemini with Google Search grounding energetic (by way of the Gemini Grounding API). Each fashions, after they determine a immediate requires present information, carry out precise Google searches behind the scenes.
This course of captures the fan-out queries as each APIs are, relatively usefully, clear about what they searched. The Gemini API returns a webSearchQueries array in the groundingMetadata discipline of each grounded response. OpenAI’s Responses API logs the precise search queries in the web_search_call output. QueryFan harvests each.
The end result is a desk: persona-specific prompts in, the precise Google search queries the AI fired out. Not what your buyer typed. What the AI looked for on their behalf. These are your new search engine optimization targets, and till now there has been no free software that surfaces them at scale.
The Grounding Query: Not Each Immediate Triggers A Search
A quick however necessary caveat before you dash off to classify the whole lot as an search engine optimization alternative.
Not every prompt causes the AI to perform a web search. The fashions decide primarily based on the consensus of token prediction as to if stay information is required.
To offer an instance, the immediate “What do crimson blood cells do?” doesn’t set off a search. The explanation is there is a really steep bell-curve of which tokens are going to seem subsequent. In the billions of coaching paperwork, the reply has stayed very steady, so an “in-model” reply can confidently be generated.
At the reverse finish of the scale, a immediate corresponding to “What occurred in the information right now?” would set off an internet search. There can be a really flat curve of “wtf tokens are subsequent?,” as there is no “steady” reply inside the coaching information; it at all times modifications, it requires stay information. It’s one other model of the Query Deserves Freshness (QDF) idea that SEOs have used for years.
In the event you’re taken with grounding, Dan Petrovic has accomplished some glorious work on this space, and even released trained models on Hugging Face to predict whether or not queries might be grounded after they hit a confidence threshold.
QueryFan surfaces which prompts triggered searches and which didn’t. Solely the grounded ones (the ones that truly brought about a Google search to occur) are actionable by search engine optimization. The in-model answers are, for now, largely outside your reach. You’d want to affect coaching information to transfer the needle there, which is a special undertaking fully, with a for much longer horizon.
What You Do With The Outcomes
You now have a listing of precise search queries that AI instruments hearth when answering questions from your particular personas. Run a normal hole evaluation:
- Which of those queries do you could have content material for?
- Which do you already rank for?
- Which have zero protection, both on your web site or wherever you’re seemingly to be talked about?
The primary two classes are diagnostic. The third is your motion checklist.
One necessary distinction from conventional search engine optimization: Your own ranking isn’t the only path to AI visibility. LLMs scan the high 10, 20, generally 50 outcomes for a grounded question and synthesize throughout them. A trusted evaluation web site rating at place 3 is a respectable route to showing in an AI-generated reply, even when your individual area by no means makes the first web page. Getting a product reviewed on a high-authority specialist web site, incomes a point out in a roundup article, showing in related group content material, all of those rely.
LLM visibility is a multi-site focus. This means the hole evaluation has two outputs: content material to create on your individual web site, and placements to earn on other people’s sites.
The Punchline
Forged your thoughts again to that Reddit quotation graph. The one which fell off a cliff when Google modified a single API parameter. A wholly impartial firm’s AI visibility tracked the habits of a search API it didn’t management and possibly didn’t know existed.
That’s the form of the dependency. And the implication isn’t that search engine optimization is lifeless; it’s virtually the reverse. search engine optimization is now working at one further take away: as a substitute of optimizing for the human question, you want to optimize for the AI-translated question that occurs between the human and Google.
QueryFan provides you a means to see what that translation truly produces. Your key phrase checklist tells you what individuals typed right into a search bar. QueryFan tells you what ChatGPT and Gemini looked for on their behalf, in the background, with out anybody asking them to announce it.
These are totally different lists. The hole between them is not a minor refinement to your content material technique. It’s the a part of AI search that no person has been measuring as a result of no person has had a free software to measure it with.
Disclosure: The writer is the creator of Queryfan.
Extra Sources:
This put up was initially printed on Mark Williams-Cook Substack.
Featured Picture: Roman Samborskyi/Shutterstock
Disclaimer: This article is sourced from external platforms. OverBeta has not independently verified the information. Readers are advised to verify details before relying on them.
