How ChatGPT, Gemini, Perplexity, Google AI Overviews and Claude Each Choose What to Recommend

The five platforms choose what to recommend in different ways. Perplexity searches the web in real time for every question and adds numbered sources. ChatGPT and Claude search the web when a question needs current information. Google AI Overviews only appear when Google's systems decide they'll help, and they draw on the Google Search index. Only some Gemini responses are grounded in Google Search results. None of them publishes its full selection rules, so the same brand often gets very different results on each platform. You have to check them one by one.

Every statement about a platform below comes from that company's official help pages. We opened each one and checked it on October 16, 2026. When we describe something we saw in our own tests that the platforms haven't stated, we say so.

Whether AI searches the web decides if it answers from old memory or from today's pages

AI answers come from two places. One is what the model learned in training, which has a cutoff date. Anthropic says each Claude model has a knowledge cutoff and may not know about events after it (Anthropic help center). The other is a web search at the moment of answering, with the answer built from the pages it finds. For a business, the difference is huge. When AI answers from memory, it may not know about the location you opened last month or the hours you changed. When it searches, your pages have a chance to be read and linked.

Here's what each platform officially says about when it searches the web:

• ChatGPT: may search the web automatically when a question would benefit from current information. Users can also choose Search from the tools menu, and people who aren't signed in can use web search too (OpenAI Help Center).
• Claude: when you ask about topics that benefit from current information, Claude uses a search tool to ground its answer in content from the live web. On Team and Enterprise plans, an owner has to turn it on first. In the new Claude experience there's no toggle, and Claude searches the web when it helps (Anthropic help center).
• Perplexity: searches the internet in real time when you ask a question, then turns what it finds into an answer (Perplexity Help Center).
• Google AI Overviews: only shown when Google's systems decide they add something beyond classic Search, and Google itself says they often don't trigger (Google: AI features and your website). AI Mode splits your question into subtopics and searches for each one at the same time (Google's AI Mode help page).
• Gemini: Google says some Gemini responses are grounded on Search results. Not all of them are (Gemini Apps Privacy Hub).

The five platforms differ in where answers come from and how they link sources

Put the official descriptions side by side and the differences jump out. The table only includes what the companies have said. Where they haven't said, we write “not public.”

Two more details make results differ from person to person. ChatGPT uses your approximate location, based on your IP address, to find local results. If memory is on, it may also put saved preferences into the search query. OpenAI's example: if you've said you're vegan and live in San Francisco, it may search for “good vegan restaurants San Francisco.” Claude may also use your location, inferred from your IP address, for local questions. So the answer you get on your own computer isn't necessarily the answer your customers see.

Being mentioned and being cited are two different things

You're “mentioned” when the AI says your name in its answer. You're “cited” when the AI links to your page as a source. One can happen without the other. Say someone asks for “kid-friendly brunch spots in Austin.” (This is a made-up example.) The AI might name a café it remembers from training without linking to the café's website. Or it might cite one café's menu page and still recommend a different place.

Keep them apart because they take different paths. Being cited has at least one clear requirement: the platform's search crawler has to be able to read your page and include it. OpenAI's, Perplexity's and Google's official docs all say so. Being mentioned has no path that clear. None of the companies explains how its model decides whom to mention.

The official reports mostly count URLs, too. Google Search Console's Generative AI performance report counts impressions of your URLs in AI Overviews and AI Mode (Google's announcement). The AI Performance report in Bing Webmaster Tools counts how often your pages were shown as sources in Copilot and AI-generated summaries in Bing (Bing's announcement). By definition, if AI only says your business name without linking to your site, neither report counts it. For ChatGPT, Perplexity and Claude, we haven't found an official performance report for site owners.

The same brand can look very different on each platform, so check each one separately

The official docs alone show how little the five platforms have in common. Each uses its own crawler and index (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot). They decide when to search at different moments, and they rewrite search queries in different ways. Even Google says AI Mode and AI Overviews may use different models and techniques, so the answers and links they show will vary.

We've tested the same brand with the same questions, and its visibility differed by more than 3x from one platform to another (test results). Even on a single platform, when we asked the same 14 questions a second time, the mention rate was 31.5% in one run and 41.7% in the other (how we measure). So don't draw conclusions from one platform or one run, and don't add the five platforms together into a single score.

To track it yourself, keep a separate record for each platform:

1. List about 10 questions customers really ask that don't include your brand name, for example “kid-friendly brunch spots in Austin.”
2. Ask the same questions in ChatGPT, Gemini, Perplexity, Google AI Overviews and Claude. Give each platform its own column, and don't merge them.
3. For each question, record three things: whether you're mentioned, whether your page is cited, and which URLs are cited.
4. Where you can, use a private window and don't sign in. Where you have to sign in, use an account with no chat history. Write down the date every time.
5. Ask each question at least twice and look at the range, not a single answer. Next month, ask the exact same questions again so you can see what changed.

No one knows exactly why AI picks one page over another

Each platform has only shared the broad strokes. OpenAI says ChatGPT ranks search results using multiple factors meant to help users find relevant, reliable information, and that placement is not guaranteed. Google says its generative AI features are rooted in its core Search ranking and quality systems. Ranking systems retrieve relevant pages, the answer is generated from those pages, and along the way the model may run several related searches at once, a technique called query fan-out (Google's guide to generative AI search). In the same guide, Google notes that no third-party tool has access to its internal ranking or AI systems. So when anyone says “AI definitely picked this page because of X,” it's a guess. That includes us.

Published research can only tell you so much. The term GEO comes from a paper first posted in 2023 and accepted to the KDD conference in 2024 (arXiv 2311.09735). The researchers built their own system. It took the top 5 sources from Google Search, had GPT-3.5 write an answer from them, and then tested 9 ways of rewriting web pages. Three methods (adding citations to sources, adding quotations from credible sources and adding statistics) improved one visibility metric by 30% to 40% in relative terms. Keyword stuffing, a common old-school SEO tactic, didn't work well. Results varied by subject area, and pages that ranked lower to begin with gained the most. The researchers also tested on Perplexity and saw visibility improve by up to 37%.

Keep this study's limits in mind. The experiments used the researchers' own system and 2023 models, not today's ChatGPT or Google AI Overviews. Treat it as a pointer that clear content with sources and numbers is more likely to be used, not as a guarantee. Our own observation is just as limited. Our website started at 3.4% AI visibility, and none of the 39 sources AI cited was on our own domain. After we made 4 changes from our report, ChatGPT cited our service page (the full story). That's one brand and one observation, and we made all 4 changes at once, so we can't tell which one made the difference.

What you can do differs from platform to platform

• For every platform: allow the search crawlers in robots.txt and in your CDN. The steps are in our website technical checklist. On the content side, Google recommends first-hand content no one else has, such as personal experience and your own point of view, rather than restating what's already online.
• ChatGPT: allow OAI-SearchBot, and make sure your host or CDN allows OpenAI's published IP addresses. If you run a restaurant, ChatGPT's reservation search may show availability from OpenTable (worldwide), Resy (US) and Yelp (US and Canada). If you're listed on one of these, that's one more way to be seen.
• Gemini: first make sure you're indexed in Google Search, since some Gemini responses are grounded on Search results. If you want Gemini Apps to be able to use your content when they answer, don't block Google-Extended.
• Google AI Overviews and AI Mode: your pages have to be indexed and eligible for a snippet, and the Search generative AI control in Search Console should stay set to include. Local businesses should keep their Google Business Profile current, and online stores their Merchant Center data. Google says both can help your products and services show up in AI responses. Track results in the Generative AI performance report.
• Perplexity: allow PerplexityBot. If you use Cloudflare or AWS WAF, follow Perplexity's official docs to add PerplexityBot and Perplexity-User to your allow rules.
• Claude: allow Claude-SearchBot and Claude-User. Anthropic says blocking them may reduce your visibility in search results (Anthropic crawler documentation).

One thing not to do: write a separate page for every way someone might phrase a question. Google says that creating content at scale mainly to manipulate rankings or generative AI responses violates its scaled content abuse policy, and that a high quantity of pages doesn't make a website higher quality.

Common questions

If you want to see where you stand on ChatGPT first, run our free quick scan. It checks ChatGPT only, needs no sign-up and takes about 1 to 2 minutes. It checks whether AI can read your website and runs 2 live questions. For the other platforms, follow the steps above and keep a separate record for each one.