ChatGPT Fell From 70% to About Half of AI Assistant Prompts. Who Are You Optimizing For Now?

Yes, ChatGPT is losing AI assistant market share in the U.S., but it is still the biggest single assistant by a wide margin. Comscore’s Q2 2026 AI Intelligence Report puts ChatGPT at about 50% of U.S. AI assistant prompts in June 2026, down from 70% in January. Gemini and Claude took most of the difference. If your AI visibility checks only ask ChatGPT, you are now looking at half the picture.

This post covers what the Comscore numbers say, what they don’t say, and how to change your tracking without throwing out work you’ve already done. It sits inside our AI search work and builds on two earlier pieces, how to track AI prompts and why AI visibility scores swing from run to run.

Is ChatGPT losing market share to Gemini and Claude?

By Comscore’s count, yes. Its share of U.S. AI assistant prompts fell from 70% in January 2026 to 50% in June, while Gemini went from 17% to 30% and Claude from 2% to 11%.

Copilot held at 5%, according to PPC Land’s September 26, 2026 write-up of the report. Smaller players moved around inside the remaining few points. Digiday’s Sara Guaglione, covering the report on September 22, described it as AI discovery “splintering” beyond ChatGPT, which is a fair read of the chart.

One caution on sources. A second outlet we checked, Marketing4eCommerce, printed Claude’s January share as 5% instead of 2%. Digiday and PPC Land agree on 2%, so that is the figure used here. We could not open the Comscore report itself, so every number in this post is as reported by those outlets.

Which AI assistant has the most users right now?

ChatGPT still does. A half share of prompts is more than Gemini, Claude and Copilot managed combined (46%), and Comscore counted 168 million ChatGPT desktop conversations in June 2026 against 89 million for Gemini, per PPC Land and Digiday.

Reach is growing for the category as a whole. Comscore found 35% of U.S. desktop users and 29% of mobile users visited an AI assistant in June 2026, up from 25% and 12% in January 2025. So ChatGPT’s share is falling inside a market that is getting bigger, and that matters. A smaller slice of a growing pie can still be more people than before.

Marketing4eCommerce’s write-up shows the same pattern on phones. ChatGPT mobile visitors went from 25 million to 47 million between June 2025 and June 2026, while Gemini went from 14 million to 19 million.

Does a share shift change which engines you should track?

It should, because a prompt set built around one assistant tells you how you look to that assistant and nobody else. When 30% of prompts go to Gemini, a blank spot there is a gap in what you know.

We aren’t saying drop ChatGPT. At 50% it deserves the most weight in any tracking set. The change is that Gemini now deserves a real second slot, Claude a smaller third one, and Copilot a watching brief because of where it sits on work machines. If your current process has one engine and everything else is “we’ll check later,” this is the quarter to fix that.

A rough starting split that mirrors the Comscore mix would put about half your prompt runs on ChatGPT, about 30% on Gemini, and the rest across Claude and Copilot. Treat that as a working default, not a rule. If your own buyers skew toward one tool, weight toward what they use.

Do you need a different strategy for each AI engine?

Mostly no. The things that get you quoted are the same on every engine: pages that can be crawled, answers stated plainly near the top, and claims that other sites repeat. Engine-specific tuning comes second.

That is the working assumption behind our earlier piece on how LLMs choose their sources, and nothing in the Comscore data argues against it. What does differ by engine is which sites get cited. Digiday’s coverage notes that Tripadvisor made up 61% of AI sourcing in the lodging category but only 21% of citations, a gap between being used as a source and being shown as one. If you work in a category like that, run the same prompt across assistants and compare who gets named.

Here is the practical split. Fix the shared basics once. Then check each engine separately for what it actually says about you, because the answers will not match.

How should you reset your prompt set after a shift like this?

Keep your existing prompts, add the missing engines, and re-baseline. Don’t rewrite the set. If you change prompts and engines at the same time you can no longer tell which one moved your numbers.

1. Freeze the current set. Record this month’s results on the engines you already track as the old baseline.
2. Add engines, same prompts. Run the identical prompts on Gemini and Claude, and on Copilot if your buyers work in Microsoft shops.
3. Run each prompt several times. Answers vary between runs, which is why a single check misleads. Our post on visibility score variance explains how many runs to average.
4. Weight the results. Roll up a blended score using your chosen engine weights, and keep the per-engine scores visible next to it.
5. Look for the outliers. A prompt where you win on ChatGPT and vanish on Gemini is a more useful finding than the average.

Comscore also reported that Claude conversations ran longer, at 9.1 prompts per conversation against 5.4 for ChatGPT and 6.2 for Gemini, according to PPC Land. Longer sessions mean follow-up questions, so test follow-up prompts on Claude and not only the first question.

How much should one quarter of panel data move your budget?

Not by much. One report is a direction, and the same report shows how fast the ground moves. PPC Land noted that Claude’s desktop conversations reached nearly 40 million in April 2026, then read 22.3 million in June, and the publication questioned how that squares with the report’s own framing. That is PPC Land’s reading, not Comscore’s, but it makes the point either way: a number that moves that much between quarters should not drive a permanent change.

A sensible rule is to move tracking coverage right away, since checking another engine costs little, and move content and ad budget only if your own referral data backs it up. Look at which assistants send you visits and leads before you shift anything that costs real money.

Comscore’s Q1 report, released June 2, 2026, showed 36% of desktop users and 23% of mobile users reaching an AI assistant. Q2 shows 35% and 29%. Desktop reach was flat between the two quarters and mobile reach rose six points, so the growth right now is on phones. Check how your pages read on a small screen, since that is where more of these answers are being read.