Is AI naming fewer brands per answer over time?

A fixed panel contains only prompts observed in every month, so growth in the number of tracked prompts does not distort the trend. On the 15,037 Google AI Mode fixed-panel prompts, the mean number of distinct brands per brand-naming answer rose from 4.8 in June 2026 to 6.0 in August, an increase of 25%. The median rose from 4 named brands to 6.

We checked the headline two independent ways. Weighting every prompt equally instead of every answer equally gives 4.5 to 5.7, also a 25% increase, and a direct recomputation from the raw records with no intermediate tables reproduces 4.8 and 6.0 exactly.

ChatGPT Search rose from a mean of 5.6 brands per brand-naming answer in June to 6.3 in July, then fell to 5.6 in August. Its net change over the quarter is +1.3%. Google AI Mode rose in July and held: 4.8, 5.9, 5.959.

The two engines also swapped places. In June, ChatGPT Search named 0.8 more brands per answer than Google AI Mode. By August, Google AI Mode named more brands per answer than ChatGPT Search, both at the mean and at the median (6 versus 5).

ChatGPT Search switched from gpt-5-5 to gpt-5-6 in our observations between August 7 and August 8, 2026. On the 16,045 panel prompts answered by both models inside August, the mean fell from 6.4 brands per brand-naming answer under gpt-5-5 (August 1 through 7) to 5.2 under gpt-5-6 (August 8 through 22), a drop of 19%. The median fell from 6 to 5.

The long list took the biggest cut. The share of answers naming 10 or more brands fell from 16.0% to 7.5%, less than half. This is the same step pattern we documented for cited sources when an earlier model update halved citation counts: engine-side changes arrive as steps, not drifts.

If something seasonal had shortened answers in mid-August, both engines would show it. Google AI Mode, which had no observed model change in August, averaged 5.9 brands per brand-naming answer on August 1 through 7 and 6.0 on August 8 through 22, a change of +0.7%.

The 19% ChatGPT Search drop therefore isolates to the gpt-5-6 update rather than to the prompt mix, the calendar, or our pipeline, all of which were shared across the two engines.

The weekly series shows the growth was not gradual. Google AI Mode averaged 4.6 brands per brand-naming answer in the week of June 29, 5.3 in the week of July 6, and 6.2 in the week of July 13. ChatGPT Search jumped from 5.6 to 6.5 in the same week of July 6.

Two engines from two companies lengthening their brand lists in the same week is consistent with both shipping answer-behavior changes on similar schedules. Google AI Mode carries no public model version label in our observations, so we can date its step but not name it.

In June, 4.4% of Google AI Mode's brand-naming answers listed 10 or more distinct brands. In July that share was 11.7%, and in August 10.3%, which is 2.3 times the June share.

ChatGPT Search moved the other way after its model update: 9.4% in June, 14.3% in July, and 10.6% in August.

The share of Google AI Mode brand-naming answers that named exactly one brand fell from 6.5% in June to 3.8% in August, a drop of 41%. ChatGPT Search held roughly flat, 4.5% to 4.5%.

For a brand that is currently the only name in an answer, this is the loss side of a longer list: sole ownership of a Google AI Mode answer became about two-fifths less common in three months.

Every bucket shifted. Between June and August, the share of Google AI Mode brand-naming answers with 2 to 4 brands fell from 44.1% to 26.3%, while the 5-to-9 bucket rose from 45.0% to 59.6% and the 10-plus bucket from 4.4% to 10.3%.

In June, half of Google AI Mode's answers named 4 or fewer brands. By August, seven in ten named 5 or more.

A market brand pool is the set of distinct brands named in any answer for one market's panel prompts over a window. Comparing matched 22-day windows (June 1 through 22 versus August 1 through 22) across the 48 markets with at least 20 panel prompts, Google AI Mode's median pool grew from 105 distinct brands to 108.5, about 3%, while its per-answer list grew 25%. ChatGPT Search's median pool went from 103 to 102.

The extra brand slots are mostly going to brands already in each market's pool, not to new entrants. Longer answers mean incumbents get named more often per answer; they do not mean many previously unnamed brands entered the pool.

Across the same matched windows, the market brand pool grew in 32 of 48 Google AI Mode markets and shrank in 15. On ChatGPT Search, 17 grew and 28 shrank.

One caution: Google AI Mode answered 42% more times per market in the August window than the June window (559 to 794 answers on average), and more answers mechanically discover more brands. So the Google AI Mode pool growth is an upper bound, which strengthens rather than weakens the conclusion of the previous insight: even measured generously, the pool grew far slower than the per-answer list.

The earlier direct-answer products, ChatGPT and Google AI Overviews, were measured over November 2025 through March 2026 on their own fixed panel of 14,272 and 14,254 prompts. ChatGPT's mean jumped from 5.7 brands per brand-naming answer in November to 8.5 in December, a one-month step of 48%, then drifted to 7.7 by March. Google AI Overviews rose from 4.7 to 6.8, up 43% over the window.

The two eras never overlap in time and used different products, so this is context, not one continuous series. But the pattern repeats: within a product generation, brands per answer trended up; the big discontinuities came from product and model changes, including the product handoff itself, which reset ChatGPT's median from 7 (March, direct answers) to 5 (June, ChatGPT Search).

We counted distinct resolved brands per answer (direct mentions only, product-line records resolved to their parent brand) across two engine eras that are never mixed in one metric: ChatGPT Search and Google AI Mode over June 1 through August 22, 2026, and the earlier ChatGPT direct answers and Google AI Overviews over November 2025 through March 2026. Every trend uses a fixed set of questions observed in every month of its era's window; movement on an unrestricted corpus would mostly measure index growth. Means and medians are over answers naming at least one brand; the share of answers naming any brand stayed between 92% and 95% in every current-era month, so extraction drift does not explain the trend.

We excluded April 2026 (a known fault dropped brand detection below 2% of answers), answers after August 22, 2026 (brand extraction was still catching up at query time), May 2026 (a partial month for the current engine pair), October 2025 (a partial ramp month for the earlier pair), and a small side channel of ChatGPT Search answers recorded without a standard model label. The legacy era's naming rate varied more (77% to 98% by month), so its monthly means carry more measurement noise than the current era's. Figures are measured in the Parse index over the stated windows.

The answer to the title question is no: over this quarter, AI answers named more brands, not fewer. That is good news for challengers only in a narrow sense. The new slots went mostly to brands already in each market's pool, and the one event that removed slots, a model update, removed 19% of them in a day.

Three practical rules follow. First, measure brands-per-answer per engine; a blended number hid a 25% Google increase and a 19% ChatGPT cut in the same window. Second, re-baseline at every model version change and treat cross-version comparisons as different products. Third, work on entering the market pool (being named at all for your market's prompts), because a longer list mostly re-names the pool it already has.