Does Perplexity use ChatGPT? Partly — and why that's not the point

The short answer: partly yes — Perplexity licenses models like GPT and Claude, but it isn't ChatGPT

Yes, partly — and the "partly" is where it gets interesting.

Perplexity is a separate company from OpenAI and Anthropic. It pays to use their frontier models — the GPT family and Claude among them — through their APIs. So when you ask Perplexity a question, a GPT or Claude model may well write the summary you end up reading.

But the model is one cog in a bigger machine. Perplexity's own plumbing — its crawler, its index of the web, its retrieval and citation system — decides which pages that model gets to read before it answers. Change the plumbing and you change the answer, even with the same model underneath.

The mix varies by mode, too. Quick Search and Pro Search behave differently, and Pro Search has at times let you pick the model yourself — GPT models, Claude, Perplexity's in-house Sonar models and others have all appeared in that picker at various times. The lineup changes often enough that any fixed list goes stale within months, and we won't pretend to a certainty the product doesn't offer.

So the honest answer is: yes, partly, and it changes. That's exactly why the consumer tech blogs ranking for this question keep going out of date — they name the models and stop there. The models are the least interesting part of the answer, especially if you own a business and want AI assistants to recommend you.

Licensing a model isn't the same as using the product

Here's the conflation worth killing: "Perplexity uses a GPT model" does not mean "Perplexity is ChatGPT".

ChatGPT is OpenAI's product. Perplexity is a competitor that buys access to some of the same models — the way a restaurant buys flour from the same wholesaler as the bakery next door. The meal still comes out different.

What Perplexity wraps around the model is a retrieval pipeline. It searches, selects sources, hands them to the model alongside your question, and shows you citations you can click. ChatGPT's search, Claude's web search and Perplexity each assemble that pipeline differently — different crawlers, different indexes, different ideas about which sources deserve trust.

The practical consequence for a business owner is the one that matters: two assistants can both be running a GPT model under the bonnet and still name different businesses, because the pages they retrieved before answering were different.

If you've been told "just make sure ChatGPT knows about you", you've been handed half a map. The other half is retrieval — and it's where most businesses are quietly losing out.

What Perplexity adds on top: its own crawler, index and citations

Perplexity runs its own web crawler, called PerplexityBot, and maintains its own index of the web. Its documentation describes its own crawler and index; it does not simply resell Google's results.

When you ask a question, the system pulls candidate pages, feeds them to the model alongside your question, and the model writes a cited answer. The quality of that answer depends on what the retrieval layer found — not on what the model memorised during training. Hold that thought; it's the hinge the rest of this article swings on.

For a business owner, the index is the part to care about. If PerplexityBot has never crawled your site — or hasn't crawled it recently — the index has nothing fresh about you to feed the model, and the answer gets assembled from whatever else it can find: directories, review sites, sometimes a competitor's pages.

One honesty note: there have been public rows between Perplexity and some publishers about how its crawler behaves. We're not going to relitigate them here. The practical takeaway for a business owner is to check your own site rather than trust anyone's summary — ours included.

Why ChatGPT, Claude and Perplexity recommend different businesses

Every AI assistant is a stack with four layers: a crawler, an index or retrieval layer, a model, and a ranking step. Change any layer and the shortlist of recommended businesses changes with it.

Different crawlers. OpenAI runs GPTBot and OAI-SearchBot. Anthropic runs ClaudeBot. Perplexity runs PerplexityBot. A site that blocks one but not the others will show up in one assistant's answers and vanish from another's — same business, same quality, different visibility.

Different freshness. An index refreshed last week knows about your new second location. A stale one is still recommending your old service area — or recommending a competitor whose new page got crawled faster.

Different local signals. Some assistants lean heavily on directories, reviews and Google Business Profile data; others lean on their own index of your pages. This is why asking ChatGPT for a recommendation produces a different answer than asking the same question in Perplexity — and why owners keep hearing conflicting advice about AI search. Both pieces of advice were probably right, for different stacks.

Picture an electrician in Leeds who has spent a year collecting mentions in trade directories. One assistant's retrieval layer rates directories highly and recommends them all over town. Another barely reads directories and works from its own crawl of the electrician's website — which is three thin pages and a contact form. Same question, two different answers, and the owner concludes AI search "doesn't work". It works. It just rewarded a different layer than the one they invested in.

The myth we're busting: "I'm in the training data, so I'm covered"

This is the myth we hear most from owners, and it's the reason this article exists.

Training data is what a model learned months or years ago, frozen at the moment of training. Retrieval is what the system reads at the moment someone asks. In our experience, for questions with a "near me" or "who should I hire" shape, retrieval does the heavy lifting.

That means two things, and both cut against the myth:

• You can be in a model's training data and still never get recommended, because the retrieval layer never fetched your pages when it mattered.
• You can be absent from every training set and still get recommended, because your pages were reachable, readable and easy to cite at the moment of the question.

So the useful question isn't "am I in the training data?" — nobody outside the labs can answer that honestly anyway. It's "can each assistant's crawler reach my site today, read my pages, and find a clear answer worth citing?"

That question is checkable. The next section is the checklist.

Four checks that decide whether AI assistants can recommend you

1. Check robots.txt — and know which bots are which

Start at yourdomain.co.uk/robots.txt and look for PerplexityBot, GPTBot, OAI-SearchBot and ClaudeBot. No mention of those names at all means they're allowed by default — absence isn't blocking. What you're hunting for is lines like this:

User-agent: GPTBot
Disallow: /

In our experience, security plugins and host firewalls sometimes block AI crawlers, and plenty of owners have no idea it has happened. Blocking is sometimes deliberate — paywalled publishers do it on purpose. For a local business it's usually an accident quietly costing you visibility.

One more wrinkle: providers often run more than one agent, and the rules differ between them. Perplexity documents a second agent, Perplexity-User, which fetches a page only when a user asks about it directly; OpenAI documents ChatGPT-User the same way. Blocking the index crawler doesn't necessarily stop that kind of one-off fetch. If blocking matters to you, read each provider's crawler documentation once rather than trusting a summary — ours included. An hour there beats a year of guessing.

2. Check your server logs

Are those crawlers actually visiting, and how often? If your hosting gives you access to raw logs, search them for the bot names above. Frequency tells you how fresh their view of your business is. A crawler that last visited in March is probably recommending a March version of your business — unless the assistant fetched the page live when someone asked, which some do.

3. Check readability

Can a machine that has never seen your site tell within seconds what you do, where you operate and why you're credible? Clear service pages, plain headings and structured data all help — structured data especially, because it gives an assistant something clean to cite. Our SEO service covers that markup work alongside the rest of the technical side.

4. Check your local signals

Assistants cross-check what they find on your site against the rest of the web. Consistent name, address and phone across directories, plus an up-to-date Google Business Profile, matter here. Our SEO and Business Profile work covers what actually moves the needle — and what's a waste of time.

If you've read our recent piece on how Google crawls and indexes pages, none of this is new thinking — it's the same crawling-and-indexing discipline applied to a bigger cast of bots. And the stakes are similar: if a crawler can't reach you, as far as that assistant is concerned you don't exist.

See what the crawlers see: run the free AI readiness check

Reading about crawlers is one thing. Knowing whether they can reach your site is another. Our AI readiness check shows whether the major AI crawlers can get to your pages and what might be blocking them — it takes a couple of minutes, needs no sign-up and assumes no technical knowledge.

Run it before you take anyone's word — including ours — about whether AI assistants can see your business. If it flags problems, our free SEO audit covers the indexing and technical side. And if you'd rather hand the whole job over, our AI search visibility service is the fix-it route for becoming the business assistants recommend.

Or book a free call and we'll walk through your results together — including whether any blocking we find is deliberate or an accident.

Questions people ask

Found this useful? Share on LinkedIn · Email it to someone