ChatGPT vs Claude vs Perplexity vs Copilot vs Google AI: What Each One Does, and What Each Lets You See

Most GEO advice treats AI engines as one thing. They are not. They crawl with different bots, draw on different indexes, show their sources differently, and give you wildly different amounts of information about whether you are being used.

That last difference decides what you can honestly claim about your own business, so I lead with it. Everything below was checked on 28 July 2026, and three of these five entries changed inside the last year, so check the dates on anything you read in this field, mine included.

The short version

Microsoft Copilot is the only engine family that hands a site owner a free citation report. Google began rolling out impression-level AI reporting in Search Console in June 2026, which is a real scoreboard and a weaker one, and not every property has it yet. Perplexity shows its sources on the page, so you can audit it by hand. ChatGPT and Claude give you crawler access and no report at all.

If you take one line from this page, I would make it this one. You can measure one of them properly, you may be able to see impressions on a second, you can audit a third by hand, and you are inferring the rest.

One more thing before I go engine by engine, because I find this one causing more damage on real sites than anything else on the list. These companies run several crawlers each, and the names are not interchangeable. A training crawler, a search-index crawler and an answer-time fetcher are three different agents doing three different jobs, and blocking one does not block the others. Most robots.txt mistakes I find on real sites come from treating one name as the switch for a whole engine.

Microsoft Copilot

Copilot leans on Bing's index, so Bingbot matters here and your Bing Webmaster Tools property is the control panel. It draws on Bing's search index plus partner surfaces running the same answer stack.

Does Microsoft Copilot show its sources?

Yes, with links in most answers. That makes it possible to sanity-check the dashboard by hand, which I have done: twenty runs on the queries Clarity said my mom's shop already wins came out at 31.8% share against Clarity's reported 33.2% for the same window. Close enough to be reassuring, and not an independent validation, because the dashboard picked the queries.

Can you measure Microsoft Copilot citations?

Yes, and this is what makes it different from every other engine on this page. Bing Webmaster Tools has an AI Performance report counting citations by day. Microsoft Clarity has an AI citations dashboard that adds the grounding queries and your share of the sources on them. Both are free and both export. If you want a number you can defend to a client or an accountant, this is where I get mine. Worth naming the incentive too: Microsoft benefits when publishers believe Copilot sends value, so I hold its numbers to the same suspicion I would hold anyone else's.

Perplexity

Perplexity runs PerplexityBot for indexing, which robots.txt governs normally. It also runs Perplexity-User, the agent behind a fetch a person specifically asked for, and Perplexity's own documentation says that one generally ignores robots.txt. So I treat robots.txt as a partial control on Perplexity rather than a complete one, and you should too before you rely on it. What it draws on is live retrieval, assembled per question.

Does Perplexity show its sources?

Yes, prominently, on nearly every answer. That single design choice makes Perplexity the cheapest competitive research in this field, and almost nobody I read uses it that way.

Can you measure Perplexity citations?

Not with a dashboard of any kind. But because it prints its sources, you can audit it yourself in twenty minutes: ask it the ten questions your customers actually ask, in a fresh incognito window, and write down which sites it cites in place of your own each time. That citation list is the engine telling you which sources it already trusts for your topic. You do not have to guess who you are competing with, and you do not need a keyword tool to find out.

ChatGPT

OpenAI splits its agents by job, and this is the one split I would memorise if you memorise nothing else here. GPTBot is the training crawler. OAI-SearchBot is what governs your appearance in ChatGPT's search results. ChatGPT-User is the live fetch made while answering someone's question. OpenAI's documentation states these settings are independent, and it gives allowing OAI-SearchBot while disallowing GPTBot as a supported example. ChatGPT draws on its training plus live browsing, and it does not browse on every answer. Check OpenAI's published list rather than mine, because it changes.

Does ChatGPT show its sources?

Sometimes, and only when it browsed for that answer. An answer written from training alone carries no citations at all, and you have no way to tell from the outside whether your business informed it.

Can you measure ChatGPT citations?

Not with a citation report, because none exists for site owners. Server logs get you further than most people assume, because those separate agents mean an answer-time fetch from ChatGPT-User looks different in your logs from a GPTBot training crawl. What no log tells you is whether the fetched page made it into the answer. So any ChatGPT visibility percentage in your inbox came from sampling. That is a legitimate method when the run count is disclosed. It is not a measurement, and it should not be priced like one.

Claude

Anthropic splits its agents the same way OpenAI does: ClaudeBot, Claude-User and Claude-SearchBot, each doing a different job in robots.txt. Claude draws on its training plus retrieval when the conversation calls for it.

Does Claude show its sources?

Yes, when it retrieved them. Same limitation as ChatGPT: an answer written from training carries nothing you can trace.

Can you measure Claude citations?

No dashboard for site owners, same as ChatGPT. Claude comes up least in the GEO content I read, which cuts both ways. There is less published guidance on it, and probably less competition for the same passages. I have no evidence either way from my own store, which is just why I am not going to tell you it is an untapped goldmine.

Google AI Overviews

This is where most GEO advice gets it wrong, mine included until recently. AI Overviews sit inside Google Search, so the controls that apply are the ordinary Search snippet directives: nosnippet, max-snippet, and data-nosnippet. Google-Extended is a different control governing Gemini apps and Vertex AI grounding, and Google's documentation states it does not affect inclusion in Search. AI Overviews draw heavily on pages that already rank for the query.

Does Google show its sources in AI Overviews?

Yes, as links inside the overview itself. Which pages get linked leans hard on what already ranks, so classic SEO is the entry fee here rather than a detour.

Can you measure Google AI Overview citations?

Not as citations, no. What changed in June 2026 is that Google began rolling out generative AI performance reporting in Search Console, showing impressions from its AI surfaces. That tells you that you appeared. It does not tell you which answer quoted you or who got named beside you, and the rollout is staged, so check your own property rather than assuming. Ordinary AI surface impressions and clicks have also long been folded into your Search totals without being broken out.

The thing all five have in common

Every one of them can only cite a page it is allowed to fetch and able to parse. That is why I keep putting the boring work first, whichever engine your customers use most. Open the right crawlers, write clean passages, keep structured data matching your visible text, and give each real question its own page. There is no engine-specific trick that beats a page that clearly answers the question.

Questions owners ask me about the different engines

Which AI engine should a small business focus on first?

The one you can measure, which today means Microsoft Copilot. That has nothing to do with user counts and everything to do with feedback: it is the only engine that will tell you whether the work is landing. Optimising for an engine you cannot measure means guessing at cause and effect on your own site for months. The foundation work is identical across all five anyway, so starting where the scoreboard is costs you nothing and teaches you fastest.

Do I need a different strategy for each AI engine?

No, and be careful with anyone selling you five strategies. The moves that get you cited are the same everywhere: allow the crawlers, answer one question per page in self-contained passages, keep your structured data matching your visible text, and get your business named in the sources those engines already trust. What differs between engines is how much they show you afterward, not what they reward.

Does blocking GPTBot stop me appearing in ChatGPT?

No, and this is the single most common robots.txt mistake in this field. GPTBot is OpenAI's training crawler. Your appearance in ChatGPT's search results is governed by OAI-SearchBot, and the live fetch during an answer comes from ChatGPT-User. OpenAI's documentation states these settings are independent and gives allowing OAI-SearchBot while disallowing GPTBot as its own example. So if you want your content kept out of training but still want to be findable inside ChatGPT, that is a supported configuration.

Does blocking GPTBot hurt my SEO in Google?

No. GPTBot is OpenAI's training crawler and has no bearing whatsoever on Google Search rankings. Google's crawlers are separate and governed separately. Blocking GPTBot is a decision about training data, and I think it is a perfectly legitimate one to make. The problem I find on real sites is rarely somebody making that choice deliberately. It is somebody who never made it and does not know the block is sitting there.

Does blocking Google-Extended keep me out of AI Overviews?

No, and this is the mix-up I see most in this field. Google-Extended governs Gemini apps and Vertex AI grounding, and Google's documentation states it does not affect inclusion in Google Search. AI Overviews are part of Search, so the levers there are the ordinary snippet controls instead. Blocking Google-Extended is still a real decision with real consequences, just not the ones people usually think they are making.

Which AI engines can I actually get a citation report from?

One family, and one partial second. Microsoft Copilot and its partner surfaces report citations through Bing Webmaster Tools' AI Performance report and Microsoft Clarity's AI citations dashboard, both free. Google reports impressions rather than citations, through the generative AI performance reporting that started rolling out in Search Console in June 2026, and not every property has it. For ChatGPT, Claude and Perplexity there is no owner-facing report at all, so anything you are shown for those three was sampled by somebody. Ask how many runs.

Is Perplexity easier to rank in than ChatGPT?

I have no evidence for that claim and neither does anyone else showing you a chart, because neither engine reports citations to site owners. What I can tell you is that Perplexity is easier to observe, since it prints its sources. Easier to observe and easier to win are different things, and the first one gets mistaken for the second constantly in this field.

How do I keep track of all five AI engines without paying for a tool?

Pick your ten real customer questions, ask them on every engine once a month in a fresh incognito window, and log which sites get cited against your own each time. Three runs per question if you can stand it, because the same query can return completely different sources minutes apart. Include Copilot in the sweep even though it has a dashboard, because it is the one column you can reconcile against a real export, which tells you whether your hand-counting is any good. Know the limit of incognito too: it clears your history but not your location, and your real customers are signed in anyway. That is a spreadsheet and an afternoon.

Where to go from here

If you want the longer argument about what each of these scoreboards is worth, and what happened when I checked the one real dashboard by hand, read which AI engines you can actually measure. To find out whether any of them can read your site at all, run the 10-minute readability check or paste your robots.txt into the free scanner.

New to all of this? Start with the plain-English GEO guide, and when you are ready to measure, the 30-minute measurement stack installs everything free. And the full case study, with its limits stated plainly, is here.