Comparisons · 14 September 2026

By Eddie Eastwood, Founder

Gemini vs ChatGPT: Which One Recommends Local Businesses?

A local business owner weighing Gemini against ChatGPT wants two answers: do they recommend the same businesses, and if not, what should they do about it? We tested this directly rather than describing how the engines are supposed to work. Both engines were run against real local-business questions (plumbers, dentists, accountants, across the UK and US) on 2026-09-14, and the transcripts below are the answers they actually gave.

What we tested, and how

We sent ten real local-business questions to ChatGPT (gpt-5.5-2026-04-23) and Gemini (gemini-2.5-pro) through the DataForSEO LLM Responses API on 2026-09-14, with web search enabled on both. Perplexity and a matching set of Google organic and local-pack results were captured alongside as controls. This is a first-party API test, not a screenshot of the consumer apps: every answer below is the model's own text, returned to a scripted question, not something anyone watched on a phone screen.

The questions covered three trades (plumbing, dental, accountancy) across UK and US locations, with one question that gave no location at all to see how each engine handled it. In nine of the ten, the city was named directly in the question text itself, so both engines had the location regardless of what the API's own location fields would accept.

Table 1: the ten questions tested

Do ChatGPT and Gemini name the same businesses?

Mostly, no. Across the ten prompts, ChatGPT and Gemini shared no cited source at all on 6 of the 10 (Leeds, the no-location kitchen-pipe question, Manchester, Austin, Bristol and Charlotte). Only the Denver emergency-dentist question produced a source both engines cited, and it's the one prompt where all three engines (ChatGPT, Gemini and Perplexity) converged on the same domain: emergencydental.com.

Table 2: shared sources between ChatGPT and Gemini, per question

Table 2 counts overlap by question. The full set of domains each engine cited across all ten questions shows how far the two engines drew on the same sources:

Even where the two engines did overlap, agreement never went very far: measured across every pair of engines and every prompt, the highest overlap between any two engines' cited domains on a single question never exceeded 27%. Both engines were working from the same question and still arrived at largely disjoint answers.

ChatGPT rank vs Google rank covers cross-engine source overlap as a broader, published third-party statistic.

How much does each one cite, and what is a citation actually worth?

The two engines disagree on which sources to use, and they cite at wildly different volumes too. Across the ten answers, ChatGPT returned between 0 and 9 citations per answer (median 4.5); across the nine answers where it actually browsed, that narrows to 3 to 9 (median 5). Gemini sat far higher, at 5 to 37 per answer (median 15). Perplexity, run as a third-engine control, was higher again and almost perfectly consistent: 19 to 20 citations on every single answer (median 20).

Perplexity returned roughly four times as many citations per answer as ChatGPT did.

But "citations" doesn't mean the same thing on every engine, and that matters for anyone trying to count how often they're mentioned. Citations returned and distinct sources are different numbers. Gemini in particular tends to cite the same grounding source more than once across a single answer: repeated annotations, not repeated new links. On the Phoenix plumbing question, Gemini returned 37 citation annotations, but those resolved to only 14 distinct URLs. If you're trying to work out whether you were cited, count the distinct domain, not the annotation total: the same source repeated five times over a long answer is one mention, not five.

Perplexity is a different case again: it returns roughly 20 sources per answer, well beyond the handful of businesses it actually names in the text. Its reference list functions more like a citation index than a shortlist.

Table 3: citations per answer, by engine

What a citation actually looks like in each engine

The two engines expose the citation itself in structurally different ways too, and that changes what "being cited" is worth to a business owner checking their own visibility.

ChatGPT places its citations as inline links inside the answer text itself, tagged with utm_source=openai: all 46 of them across the ten answers carry that tag, including one that points to a Google Maps address rather than a business's own page. Click one and you arrive at the destination with that tracking parameter attached.

Gemini works differently. Every one of its 165 citation URLs across the ten answers is a Google Vertex AI grounding redirect, not a direct link to the source. The real domain only appears in the citation's title text, not in the URL a reader would click. That is the documented mechanism, not a quirk of this test: Google's own Gemini API documentation describes grounded answers as returning "inline annotations directly on the text content block," which "provide citation information linking parts of the response to their sources": annotations, in other words, rather than direct outbound links.

For a business owner, the difference is practical: a ChatGPT citation is a link a reader can click straight through to your site. A Gemini citation is a name attached to a redirect wrapper: the attribution exists, but the direct traffic path does not work the same way. How AI engines source differently covers the retrieval side of this in more depth; what matters here is the outcome each mechanism produces for a cited business.

What happens when the buyer doesn't say where they are

One of the ten questions gave no location at all: "My kitchen pipe is leaking badly right now. Who should I call?" The three engines handled it three different ways.

ChatGPT declined to browse for that question at all and asked the user directly for a city or ZIP code, naming no business. Gemini did browse, but also asked for a location before naming anyone. Perplexity did neither: it answered with three named plumbing firms in Baton Rouge, a city that appears nowhere in the question.

This is one observation on one prompt on one date, and the inferred city is more consistent with the calling infrastructure's own network location than with any real user's location, so it should be read as a single behaviour, not a rate. It is still worth knowing before you assume "no location mentioned" means "no answer given": one of the three engines tested here answered anyway.

Location handling wasn't the only single-prompt divergence worth flagging. On the Leeds plumbing question, ChatGPT was the only one of the three engines that declined to name a plumber directly as its own pick, routing instead to vetted directories (Checkatrade, Which? Trusted Traders, Yell) and naming only one trading business, quoted from within a directory listing. On the other nine questions it named businesses directly, so this is a single-prompt behaviour, not a general ChatGPT pattern.

How this compares with what Google itself shows

Google's own results for six keyword-shaped versions of these local questions returned a local pack on every one, and an AI Overview on none of the four we specifically tested for it. Two unrelated control queries run on the same endpoint on the same day did return an AI Overview, which confirms the absence wasn't a broken test: it's what Google actually served for these particular queries, that day. The sample is narrow (four queries, one date, one datacentre IP), so this is not a claim that Google never shows AI Overviews for local searches, just what happened on this specific run.

Where the AI engines' citations overlapped with Google's own organic rankings for the same query, the picture was inconsistent rather than clean: anywhere from none of an engine's cited domains appearing in Google's top results, up to 8 of 15 for Perplexity on the Leeds question. There's no single number that summarises this well, and one of the six comparisons used a partial Google result set that isn't comparable to the rest.

Table 4: local pack vs AI Overview presence, six queries tested

What this means for your business on Monday morning

The instruction that actually follows from this data is narrower than "optimise for AI search": being recommended by one engine tells you nothing about the others. If you've only ever checked ChatGPT, or only ever checked Gemini, you have information about exactly one surface.

In practice: ask your own real buying questions, the way a customer would actually type them rather than a generic "best [trade] near me", into ChatGPT and Gemini separately, on the same day, and look at what each one names and cites. Don't assume a result from one carries over to the other; the data above says it usually doesn't.

Why a weaker competitor sometimes gets named ahead of a stronger one is covered in why AI names worse businesses, and the full set of levers for improving your own visibility across engines is in the local AEO guide.

If you'd rather see where your own business currently stands across these engines than run the check yourself, QBiz's AI visibility audit does exactly that comparison for your business name.

Frequently asked questions

Do ChatGPT and Gemini recommend the same local businesses?

Not usually. Across ten local-business questions tested on the same day, ChatGPT and Gemini cited no source in common on 6 of the 10. Only one question produced a source all three tested engines cited.

Why does ChatGPT cite fewer sources than Gemini?

The two engines return citations differently rather than one simply finding less. ChatGPT returned a median of 4.5 citations per answer across our test set; Gemini returned a median of 15. Part of the gap is structural: Gemini can cite the same grounding source more than once across an answer, while ChatGPT's citations are inline links in the answer text.

Is a Gemini citation the same as a website link?

No. Every citation URL Gemini returned in this test was a Google Vertex AI grounding redirect, not a direct link to the business's site: the real domain appears only in the citation's title, not the clickable URL. ChatGPT's citations are inline links directly in the answer text.

Does Google show an AI Overview for local searches?

Not consistently, at least on the queries we tested. Six local queries all returned a Google local pack; none of the four we tested for an AI Overview returned one, while two unrelated control queries run the same day did. This is a small, dated sample, not a general rule about local search.

What should a local business actually do with this?

Check ChatGPT and Gemini separately, using your own real buying questions, rather than assuming a result on one applies to the other. The data above shows the two engines mostly draw on different sources for the same question.