Do AI Engines Name the Same Top Brand? What the Data Shows

AI engines disagree on the top brand more often than most teams assume. In a sample of 5,810 buying topics, Bluefish reported that ChatGPT and Google put a different brand first in 35% of them. That is two engines and three categories. Add Gemini, Perplexity and Claude and the possible disagreements multiply, which is why one engine cannot stand in for the rest.

Key takeaways

• The only published top-brand disagreement rate we found is Bluefish's 35% for ChatGPT versus Google. It is company-reported and the full method is not public.
• Ahrefs measures something different: how stable citations are over 20 days. Its finding that one check misleads applies to every engine comparison too.
• Define "top brand" before you count: first brand named, most-named brand, or first brand recommended. Each gives a different rate.
• With 20 prompts, an observed 35% disagreement rate carries a margin of roughly 21 points either way. Use 100 or more prompts before you act on a gap.
• Track each engine separately. A blended "AI visibility" score hides exactly the disagreement you need to see.

What this post does and does not contain

This was meant to be a report on Noma by Nurdd's own daily prompt runs across its engines. This edition does not include that dataset. We have no vetted, anonymised top-brand disagreement figures from it to publish, and we will not put a number here that the data does not support. There are no Noma charts for the same reason.

What you get instead is the published evidence, a close reading of how its methods differ, the arithmetic on sample size, a worked grid you can copy, a decision table for what to do with the result, and a test you can run on your own category today. Noma's one published count is on a different question: its own analysis of 2,545 ChatGPT, Gemini, Perplexity and Claude answers looked at how often engines named Nurdd, not at which engine disagreed with which.

Do ChatGPT and Gemini recommend the same brands?

Not reliably. Bluefish, an AI marketing platform vendor, said on 27 October 2026 that ChatGPT and Google put a different brand at the top of AI-generated answers in 35% of 5,810 buying topics (PPC Land, October 2026). Its own announcement describes the sample as 5,810 buying topics across 3 categories (PR Newswire, October 2026). Bluefish repeats the finding on its own blog, where it argues for platform-specific strategies (Bluefish, October 2026), and eMarketer, October 2026, repeats the same figure.

Read that as a signal, not a law. ContentGrip, October 2026, notes the figure describes a sample of topics, not every category or shopper. PPC Land adds that Bluefish has not published the details of the research. So the comparison unit (topics, not prompts), the definition of "leads" and the number of runs per topic are unknown to us.

One consequence is easy to miss. The headline compares ChatGPT with Google. Google is not one product: AI Overviews and AI Mode are separate surfaces. The sources we found do not say which Google surface Bluefish used, so the result should not be transferred to a ChatGPT versus Gemini pair, a ChatGPT versus Claude pair or a Perplexity pair.

How often do AI engines disagree on the top brand for a topic?

The honest answer is that it depends on four choices, and the published 35% fixes some of them without showing us which.

The last row is an expectation, not a finding. Treat it as a hypothesis to test in your own grid.

Pairs, not just engines

With two engines there is one pair to compare. With four there are six: ChatGPT-Gemini, ChatGPT-Perplexity, ChatGPT-Claude, Gemini-Perplexity, Gemini-Claude and Perplexity-Claude. Thirteen engines would give 78 pairs. A single "disagreement rate" therefore means little until it says which pairs it averages and whether it requires all engines to agree. All-engine agreement is always at or below the weakest pair, so it is the harshest test.

Is there data on how consistent AI answers are over time?

Yes, for citations. Ahrefs published its study on 29 October 2026. It followed cited websites across five AI tools for 20 days and reported that 50–83% of citations are gone the next day, but about 75% of those that disappeared returned within the 20-day study (Ahrefs, October 2026). Its takeaway, in its own words: "Don't judge AI visibility from one check."

The Ahrefs tables show how different the tools are from each other, which matters for any cross-engine claim. In its data, the share of answers with no source citations at all was 64% for Gemini and 53% for Copilot, against 0.7% for ChatGPT and 0.9% for Google AI Overviews. The share of citations gone by day two ranged from 50% on AI Overviews to 83% on Copilot. Engines do not just choose different brands. They differ in how much evidence they show at all.

Ahrefs also looked inside ChatGPT. It reported that ChatGPT repeated exactly the same fan-out searches as the previous day only 26% of the time, and that cited websites changed 81% of the time even when the searches stayed the same, against 91% when they changed. That suggests much of the churn happens after retrieval, when the engine decides what to cite. For a brand, the implication is direct: an engine disagreeing with itself is a normal condition, not a fault.

How do the Bluefish and Ahrefs studies compare?

They answer different questions, so neither validates the other's number. Put side by side:

Neither includes Claude or Perplexity as a head-to-head pair on top brand, as far as these sources show. That gap is the reason to run your own comparison.

Why does ChatGPT name a different brand than Perplexity for the same question?

Four mechanisms are enough to explain most splits. We list them as working explanations, not as findings from the studies above.

1. Different retrieval. Engines send different searches and read different pages. Ahrefs' ChatGPT figure shows even one engine changes its searches most days.
2. Different weighting. Given the same pages, engines pick different cited sources, as the 81% figure above hints.
3. Search versus memory. Some answers lean on live results, others on what the model learned in training. A brand that is strong in older coverage can lead in one and trail in the other.
4. Different answer shapes. An engine that lists five options and one that names a single pick will not produce the same "first brand", even from identical evidence.

Our earlier explainer on why ChatGPT, Gemini, Claude and Perplexity disagree about your brand covers the mechanisms in more depth, and how AI engines decide which sources to cite covers the source layer.

How should you define "top brand" so the number means something?

Choose one definition before you collect data and keep it for every later run. Changing it mid-way breaks the trend line.

For a first pass, use "most-named brand across three to five runs". It separates real engine differences from one-off noise better than a single run.

How many prompts do you need before a gap is real?

Treat a disagreement rate as a proportion and give it a margin. Using the normal approximation at 95% confidence, the margin is 1.96 × √(p × (1 − p) / n).

These are our own calculations from the formula, not measured data. They assume prompts are independent, which they rarely are, so real margins are wider. A 20-prompt spot check can tell you engines differ, but not by how much.

How do I check which brand each AI engine names first for my category?

1. Write 20 to 50 unbranded buyer prompts. Use the questions a buyer asks before they know your name. The Free AI Prompt Finder surfaces candidates, and which prompts to track explains how to keep the set stable.
2. Fix your definition of top brand. Use the table above and write it down.
3. Run each prompt three to five times per engine, in fresh sessions, on ChatGPT, Gemini, Perplexity and Claude. Record the first brand each time and take the most frequent as that engine's answer.
4. Fill in a grid with prompts as rows and engines as columns.
5. Count disagreement two ways: pairwise across the six pairs, and all-four agreement.
6. Report the denominator and the margin, as in the table above.

The free Free Competitor AI Share of Voice tool compares how often ChatGPT, Gemini, Claude and Perplexity name you against up to three competitors, and shows which sources they cite. It is a quick way to see whether the engines already split on your category before you build a full prompt set.

A worked grid (hypothetical numbers)

The grid below uses invented brands A, B and C and five invented prompts. It exists only to show the arithmetic. It is not Noma data and not a result about any real category.

All four engines agree on 2 of 5 prompts (1 and 5), so all-four agreement is 40% and all-four disagreement is 60%. Now count pairs. Each prompt has six pairs, so there are 30 pair checks. Prompt 1 has 0 disagreeing pairs. Prompt 2 has 3 (Gemini against each of the others). Prompt 3 has 3 (Perplexity against each). Prompt 4 has 4 (the A pair against the C pair, 2 × 2). Prompt 5 has 0. That is 10 of 30, a pairwise disagreement rate of 33%.

Same grid, two numbers: 60% and 33%. Neither is wrong, and a report that does not say which one it uses cannot be compared with anyone else's, including Bluefish's 35%.

What should you do with the result?

The pattern in your grid tells you where to spend effort. This is our reasoning framework, not a measured rule.

What this means for how you track

• Report per engine, not blended. If ChatGPT names you first and Claude never does, an average of 50% describes neither.
• Use repeat runs. AI citation stability shows why one answer is a sample, not a result.
• Separate mentions from citations. Ahrefs found that in ChatGPT, 54% of cited websites were not named in the answer text, and 13% of next-day cases lost the link but kept the name. See citation vs brand mention.
• Prioritise by where buyers are. Multi-engine AEO covers how to rank engines when they split.

Limits of this evidence

• The Bluefish figure is company-reported, covers two engines and three categories, and its details are unpublished per PPC Land. Bluefish also sells AI marketing software.
• The Ahrefs study measures citations, not top-brand agreement. We use it only to show run-to-run instability and structural differences between tools.
• The two studies use different units, engines and periods, so we do not combine their numbers into one rate.
• Our margin table assumes independent prompts and a simple proportion. It is a guide, not a test.
• The worked grid is invented for arithmetic.
• This post contains no Noma disagreement data. Do not read any number here as a Noma result on engine disagreement.

Where Noma fits

Noma by Nurdd is an AEO platform that asks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, the questions your buyers ask, every day. It reports visibility, position and sentiment per engine, shows where competitors are named instead, and lists the sources each answer cites, so engine-by-engine differences are visible instead of averaged. See it on your own brand: book a demo or start the 3-day free trial. A card is taken at signup so billing can start on day 3; cancel before then and nothing is charged. The prompt tracking page explains the fixed-set method.

How we researched this

We reviewed the vendors' own announcements (Bluefish's press release and blog), trade coverage of the Bluefish research (PPC Land, ContentGrip, eMarketer) and Ahrefs' published study with its method section and tables. Facts and figures were checked on 2026-11-03. We did not find Bluefish's full methodology, so we have not described it beyond what those sources state.

What to do next

Run the six-step check above on 20 prompts this week, then decide whether you need a larger, daily set. To see per-engine top brands for your category without building the grid by hand, book a demo or start the free trial.