How does AI decide what to recommend?
What ChatGPT, Claude, Gemini and Perplexity draw on when they name a business: training memory, live search, the pages they read, and why the answer is not a ranking.
Short answer. An AI assistant recommends a business from two places: what it learned in training, which stops at a cut-off date, and what it finds by searching the web when you ask. Search is the part that changes week to week. The result is a written answer, not a ranked list, and it can differ each time.
The two routes
Every recommendation comes from one of two routes, and often both.
| Training (memory) | Live search | |
|---|---|---|
| Where it comes from | A snapshot of text the model was trained on | Web pages read at the moment of the question |
| How current | Stops at a cut-off date | As current as the search index behind it |
| When it changes | When a new model is released | Whenever the pages or the index change |
| Can you see the source? | Rarely | Usually, as citations or a source list |
Anthropic, for example, lists a June 2026 knowledge cut-off for its current Claude models (Anthropic). Anything after that date reaches Claude only through search. The full explanation is in does ChatGPT know my business, or does it search for it.
When does an AI search instead of answering from memory?
The assistant makes this choice itself. Anthropic's documentation says Claude searches when a request depends on information that is current, changing or outside its training data, and answers directly when it draws on stable knowledge (Anthropic). Google says Gemini's model "determines if a Google Search can improve the answer" and, if so, generates one or more search queries and runs them (Google AI for Developers).
OpenAI's help centre says ChatGPT will search automatically when a question might benefit from information on the web, and offers a manual Search option (OpenAI help centre). A question such as "best dentist near me" depends on current, local information, so it is the kind of question that tends to trigger a search.
What happens during a live search
- The assistant rewrites your question into search queries. OpenAI says it may send rewritten queries to search providers (OpenAI help centre). Google says AI Overviews and AI Mode may use "query fan-out", running several related searches across subtopics (Google Search Central).
- A search index returns pages. Which index depends on the assistant (next section).
- The model reads the results and writes an answer. Perplexity describes this as interpreting the question, searching in real time, and compiling the most relevant information with numbered citations (Perplexity).
- Sources are shown. Claude's documentation says citations are always enabled for web search (Anthropic).
Because the assistant may run several searches, your business can be missed by the first query and found by the second. That is one reason a single check proves little.
Which search index each AI reads
A page the AI cannot find in its index cannot be recommended from live search. The evidence for each row, with dates and gaps, is in which search engine each AI uses.
| AI | Live search draws on | How sure we are |
|---|---|---|
| ChatGPT | Several providers plus OpenAI's own crawler. OpenAI has not said how much comes from Bing | Partly confirmed (OpenAI) |
| Claude | Brave Search, according to Anthropic's subprocessor list | Reported, not described in Anthropic's documentation (TechCrunch) |
| Gemini and Google AI features | Google Search's index | Confirmed (Google) |
| Perplexity | Its own index, built by its crawler | Confirmed (Perplexity) |
What makes a page eligible
The rules are plainer than most advice suggests.
- Be indexed. For Google's AI features, a page must be indexed and eligible to appear in Search with a snippet. Google says there are no additional technical requirements and that no special markup or machine-readable file is needed (Google Search Central).
- Let the AI's crawler in. OpenAI says sites opted out of
OAI-SearchBotwill not be shown in ChatGPT search answers (OpenAI). Perplexity says to allowPerplexityBotto appear in its results (Perplexity). Anthropic says blockingClaude-SearchBotmay reduce visibility in search results (Anthropic). See what the AI crawlers do. - Match the question. The assistant reads pages to answer a specific question. A page that states plainly what you do, where, and for whom gives it something to use. Google's documentation does not publish a ranking formula for AI answers, and neither does any other provider we checked, so treat advice that claims to know the exact weights with caution.
Why a competitor appears and you do not
These are the usual causes, from most to least checkable:
- Your site is blocked or not indexed in the index that assistant uses.
- The competitor has pages that answer the question directly, while yours describe the business in general terms.
- The competitor is mentioned on other pages the AI reads, such as directories, reviews or articles. Open the sources the AI lists to see which pages it relied on.
- The model's memory favours them. A business widely written about before the cut-off may be named without any search.
None of these is a score you can read off a dashboard. The first is something you can confirm in minutes, and the third is visible in the AI's own source list. How to check if AI recommends your business shows the method.
It is not a ranking, and it changes
An AI answer is written fresh each time. A SparkToro and Gumshoe.ai study ran 2,961 prompts across ChatGPT, Claude and Google's AI features and found the same question rarely produced the same list of brands, with the odds of an identical list under 1 in 100 (Search Engine Land). The study is not peer-reviewed and was done with a company that sells AI tracking tools, so read it as a strong hint, not a measurement of every market. It does support one practical rule: measure how often you appear over repeated runs, not whether you appeared once. More in why AI answers change every time.
Where to go next
- Check your own position: how to check if AI recommends your business, then one guide per assistant: ChatGPT, Perplexity, Claude and Gemini.
- Understand the pieces: memory versus live search, which search engine each AI uses, what the AI crawlers do and where AI assistants get their sources.
- Act on it: how to get recommended by ChatGPT.
Frequently asked questions
Is there a ranking list I can climb to be recommended by AI? No single list exists. A study of 2,961 prompts found the same question rarely returned the same list of brands twice, so AI recommendations behave more like a probability than a position. What you can measure is how often you appear across repeated runs.
Does paying an AI company get my business recommended? None of the documentation we read describes a way to pay for a place in an AI's organic answers. Google says there are no extra requirements beyond being indexed and eligible for Search with a snippet to appear in its AI features (Google Search Central).
Do all AI assistants use the same sources? No. Gemini and Google's AI features draw on Google Search. Perplexity runs its own index. Claude's search is reported to use Brave Search. ChatGPT uses several providers and OpenAI has not said how much comes from each. A page can be found by one and missed by another.
How long before a change on my website shows up in AI answers? It depends on the route. Live search can pick up a new page once the search engine behind it has indexed it, which can take days. Training memory only changes when a new model is released with a later cut-off.
Can an AI recommend a business it has never heard of? Yes, if live search is on and the business's pages are indexed and match the question. With search off, the model answers from memory only, and a business that was rarely written about before the cut-off may not appear.
Sources
Every source above was checked on 27 October 2026. The index-by-AI table, with its gaps, is maintained at which search engine each AI uses.