ChatGPT Plays Favorites With Brands. Gemini Mostly Doesn’t. Here’s Why That Matters for Your GEO Strategy

Ask ChatGPT which project management tool to use, and there’s a good chance it’ll just tell you: “I’d go with X.” Ask Gemini the same question, and you’re more likely to get a tidy list of four options with no clear winner.

That’s not your imagination. Google AI Overviews are the AI-generated answer summaries Google shows at the top of some search results. A September 2026 academic audit study put numbers to exactly this gap — how ChatGPT, Gemini, and Google AI Overviews talk about brands and products when someone asks for a recommendation. The results are worth sitting with. They change how “AI visibility” should be thought about. It isn’t one thing — it’s several, platform by platform.

One query, three completely different answers

A lot of GEO advice treats “getting recommended by AI” as a single skill. It isn’t. Some AI tools will tell you flat-out which brand to pick. Others lay out several options and leave the decision to you. That’s not a small stylistic quirk. It’s a fundamentally different way of answering the same question — and it changes what “winning” on a given platform actually looks like.

GEO — Generative Engine Optimization — is the work of getting a brand mentioned or recommended by AI tools like ChatGPT and Gemini, the way SEO is about getting found on Google. Most brands today run GEO across more than one AI platform, and for them, that difference is the whole ballgame. Treating ChatGPT, Gemini, and Google AI Overviews as interchangeable means missing how differently each one actually behaves.

The numbers: 79% vs. 7% vs. 2%

The audit study looked at how often each platform’s product recommendations included a clear, first-person statement of preference — something like “I recommend X” — rather than a neutral list with no stated favorite.

The gap is not subtle:

Ask ChatGPT for a recommendation, and it behaves like an opinionated friend. Ask Gemini or Google AI Overviews, and you get something closer to a reference desk clerk — options handed over, no opinion attached.

One thing worth being precise about: this stat isn’t measuring accuracy, and it isn’t measuring how often your brand gets mentioned. It’s measuring tone. “Brand X is a popular choice for this” is a neutral mention. “I recommend Brand X” is a stated preference. ChatGPT leans hard toward the second. Gemini and AI Overviews lean hard toward the first.

Why should a marketer care about tone specifically? Because a stated preference carries more weight with the person reading it. Being the brand ChatGPT says “I recommend” is a different probably more valuable outcome than being one of four names Gemini lists without comment.

The bigger issue: they’re not even reading the same web pages

Here’s the finding that matters more than the tone difference, and it’s easy to miss if you stop at the first-person stat.

For the same query, the ChatGPT and Gemini consumer interfaces — the websites and products people actually use — pulled answers from the same web domains in only 5.4% of cases. Put plainly: ask both platforms the identical question, and roughly 95% of the time, they’re drawing from different sources on the web entirely. Not the same page phrased two ways — often a completely different page.

This is the number that should actually reshape your GEO checklist. If two AI platforms mostly aren’t looking at the same sources for the same question, being optimized for one platform’s apparent source preferences tells you almost nothing about the other. A page ChatGPT cites constantly might be a page Gemini has barely touched for the same topic, and vice versa. “Getting cited by AI” isn’t one goal. It’s several — one per platform.

Why testing the API isn’t the same as testing the app (and why that’s a real problem)

Quick definitions first, because this trips people up outside the technical weeds. The “app” is the actual product you or your customers open and type into — the ChatGPT website, the Gemini app. The “API” is a separate, developer-facing way of sending the same kind of request programmatically — usually how third-party tools automate checks at scale, rather than a person typing a question by hand.

In theory, a company’s API and its consumer app should behave the same way. In practice, according to this study, they often didn’t.

The ChatGPT API and the Gemini API each matched their own companies’ consumer apps poorly. Testing ChatGPT’s API, in other words, did not reliably predict what a real person would see typing the same question into the actual ChatGPT app. Same story for Gemini.

That’s a quiet but serious problem for the entire AI-visibility audit industry. A lot of tools claiming to measure “how visible is my brand on ChatGPT” run their checks through the API — it’s cheaper, faster, and easier to automate at scale. If the API doesn’t reliably match what real users see in the app, those tools may be handing you a number with little connection to your actual visibility to actual customers. The study’s source material is blunt about this: audits need to test the apps people actually use, not just the APIs sitting underneath them.

So what do you actually do differently?

Stop treating “GEO” as one undifferentiated channel. If ChatGPT, Gemini, and Google AI Overviews differ this much — in tone, in sourcing, in app-vs-API behavior — a single “GEO strategy” built on the assumption of uniform behavior is working off the wrong model. Think of it closer to running separate channel strategies, the way paid search and paid social need different playbooks, rather than one unified “AI SEO” push.

Platform by platform, that looks like this:

  • ChatGpt. Given its tendency to state a clear preference, getting positioned as the recommended option matters more here than simply being mentioned somewhere in the answer. This is closer to a “win the recommendation” game than a visibility game.
  • Gemini. With far less first-person preference-stating, the goal shifts toward being reliably present across the neutral list of options Gemini surfaces — and understanding which sources Gemini actually pulls from for your topic, since per the domain-overlap finding, that may be an entirely different set of sources than what works for ChatGPT.
  • Google AI Overviews. The most neutral of the three, and likely the most tied to existing Google Search ranking signals and the broader Google ecosystem. Classic SEO fundamentals still carry real weight here.

And the testing part matters just as much as the strategy part: if you or a vendor are running AI-visibility audits purely through API calls, treat the results as a partial picture, not the full one. Wherever you can, test the actual consumer-facing apps — the real ChatGPT app, the real Gemini app — the way an actual customer would use them.

This kind of platform-by-platform variance isn’t theoretical for us, either. “ChatGPT visibility,” here, means how often ChatGPT mentioned a brand at all in answers relevant to it. Across more than 60 of our own brand-tracking projects, that average came out to 3.5% for BFSI brands (banks, NBFCs, insurance, and fintech companies), 5.6% for healthcare brands, and 7.4% for education brands (a smaller, more directional sample there). Those numbers are ChatGPT-specific.

Given what this study shows about how differently Gemini and AI Overviews behave, there’s no good reason to assume those same percentages — or even the same ranking between industries — would hold on another platform. Assuming one platform’s numbers generalize to the others is exactly the trap this study exposes.

A checklist before your next multi-platform GEO audit

  • Test each AI platform separately. A ChatGPT result doesn’t predict a Gemini result, or vice versa.
  • Where possible, test through the actual consumer app — not just an API call — before trusting any vendor’s “AI visibility score.”
  • Track whether your brand is being mentioned versus actively recommended. These are different outcomes, especially on ChatGPT.
  • Check which domains each platform is actually citing for your key queries. Don’t assume the sources that work for one platform are landing on another.
  • Revisit your audit methodology periodically. This space moves fast, and so does the tooling measuring it.

FAQs

What does it mean when ChatGPT “recommends” a brand?

It means ChatGPT’s answer includes a clear, first-person statement favoring one brand or product over the alternatives — “I’d recommend X,” not just a mention among several options.

Why does Gemini rarely say “I recommend” a specific product?

Based on this audit study, Gemini’s product-recommendation answers are far more likely to present a neutral list without a stated favorite — about 7% of the time, versus ChatGPT’s 79%.

Is Google AI Overviews safer for brands than ChatGPT?

It behaves more neutrally — a stated preference in only about 2% of cases here — but “safer” depends on your goal. Neutral mentions still require your brand to show up and get cited; the chance to be the single clear “recommended” answer is just smaller on AI Overviews than on ChatGPT.

What’s the difference between testing an AI platform’s API and testing its app?

The app is the product people actually use directly — opening ChatGPT and typing a question. The API is a separate, developer-facing way other tools query the same system programmatically. This study found the two didn’t reliably match for either ChatGPT or Gemini, so audits run only through an API may not reflect what real users see.

Does this mean I need a different GEO strategy for every AI platform?

Directionally, yes. The platforms differ enough in tone and in which sources they draw from — only 5.4% domain overlap between ChatGPT and Gemini for the same query here — that a one-size-fits-all GEO approach will likely underperform on at least one of them.

The bottom line

ChatGPT, Gemini, and Google AI Overviews aren’t three doors into the same room. They pull from different sources, talk about brands differently, and per this study — don’t even test consistently between their own apps and APIs. Treat “AI visibility” as one number, and you’re very likely missing a real gap on at least one of these platforms. The fix isn’t more GEO effort. It’s platform-specific GEO effort, tested the way people actually use these tools.

If you’re trying to figure out where your brand stands on ChatGPT specifically versus Gemini, or want a proper multi-platform GEO audit that tests real consumer apps instead of just APIs — that’s exactly the gap we help close.

Shahrukh Saifi

Shahrukh Saifi Home Shahrukh Saifi Shahrukh Saifi Linkedin Our Mission & Vision Executive Profile A highly accomplished and data-driven executive with over 18 years of...