New brands struggle to reach #1 in AI recommendations

A #1 pick is the brand an AI answer names first. Early brands are the 44,030 brands AI named in the first four weeks of our tracking, October 3 to 31, 2025. In September 2026, early brands took 64.4% of ChatGPT Search #1 picks (103,778 of 161,244 answers) and 66.8% of Google AI Mode #1 picks (105,021 of 157,265).

Early brands are 6.0% of the 731,975 brands AI named over the whole year. So about 1 in 17 brands took about 2 in 3 of the top slots, one year, one product change, and four ChatGPT model versions later.

We checked the headline two ways. On the fixed panel (prompts answered in all nine measured months) the shares are 63.1% and 66.6%. A recount from the separate brand-mention records gives 64.36% and 66.78%, matching to within 3 answers.

In September 2026, 28,856 brands were the #1 pick in at least one ChatGPT Search answer. Early brands were 9,454 of them, or 32.8%, and they took 64.4% of the picks. On Google AI Mode, early brands were 10,550 of 25,437 #1 brands (41.5%) and took 66.8% of the picks.

Brands AI first named in April 2026 or later took 11.4% of ChatGPT Search #1 picks and 7.6% of Google AI Mode #1 picks. Google AI Mode leaves less room for newcomers at the top than ChatGPT Search.

AI first named 182,517 brands between April 6 and October 6, 2026. 14.8% of them were the #1 pick in at least one answer. For brands first named between November 2025 and April 5, 2026 the share is 25.8%, and for early brands it is 74.3%.

Most newly named brands appear in only a handful of answers. Among brands with at least 5 answers, 47.3% of the newest group reached #1, the same as the November to April group (47.2%). Early brands with 5+ answers reached #1 at 79.8%. The gap is between early brands and everyone after them, not between this year's newcomers and last year's.

We followed the 179,787 brands AI first named between January 1 and March 31, 2026, for at least 189 days each. 23.4% were in the top three of an answer on the day AI first named them. Another 15.9% reached the top three later: the median wait was 40 days and 1 in 10 waited 194 days or more. The other 60.7% never reached the top three.

Discovery lag is the number of days from the first answer that names a brand to the first answer that puts it in the top three. It has two modes. A brand usually enters through a question narrow enough that it fits at the top, or it enters the list low and climbs slowly.

Discovery lag varies by market. Among brands that reached the top three after their first day, the median discovery lag was 11 days in Marketing & Client Acquisition Tools and 35 days in CRM and Sales Pipeline Software. Product Analytics and Session Replay (34 days) and Digital Adoption & Interactive Demo Platforms (30 days) are also slow.

In CRM and Sales Pipeline Software, the slowest market, 7.5% of brands AI first named there in the last six months became the #1 pick in at least one answer there. In Business VoIP Phone Systems, one of the fastest, the share was 25.3%.

We compared answers to the same prompt on consecutive days. From June to September 2026, ChatGPT Search named the same #1 brand 50.0% of the time and Google AI Mode 50.4%. The overlap of the full brand lists (shared brands divided by all brands named on either day) averaged 0.38 and 0.40.

One day's answer says little. The early-brand advantage in this study only shows up because we count thousands of answers per market per month. It also means every change in this article has to be measured against this day-to-day movement, which we do below with a noise threshold.

The concentration score is the Herfindahl-Hirschman index of a market's #1 picks, from 0 to 10,000. A score of 1,000 is what ten brands splitting the #1 picks equally would produce; higher means fewer brands share them. We corrected it for sample size, because markets got fewer answers per month after the switch to ChatGPT Search and Google AI Mode.

In the median of the 50 largest markets, the score went from 880 in November 2025 to 932 in September 2026 on the ChatGPT side, and from 964 to 1,009 on the Google side. The top three brands in the median market held between 42% and 50% of #1 picks in every measured month. Concentration moved, but it did not trend toward fewer brands.

For each market we split every prompt's answers in a month into two halves and measured how far the two halves' concentration scores differ. That sets a noise threshold: a monthly change bigger than it is unlikely to come from re-asking alone. At this threshold about 5% of moves would cross by chance.

Across 700 market-month moves within one engine generation (50 markets, 7 month-to-month steps, 2 engines), 65.7% stayed within the noise threshold. 19.1% concentrated beyond it and 15.1% spread beyond it. Real changes happen, and they go both ways at similar rates.

ChatGPT Search answers switched from gpt-5-5 to gpt-5-6 on August 8, 2026. Over the six weeks before, the weekly concentration score in the median large market averaged 986. Over the six weeks after, it averaged 886, a change of -10.1%. Google AI Mode, which had no model change, moved +2.6% over the same weeks (916 to 939).

Market by market, from July to August 16 of 50 markets spread beyond the noise threshold on ChatGPT Search and 6 concentrated. On Google AI Mode it was 6 and 13. The earlier change from gpt-5-2 to gpt-5-3 in March 2026 shows no clear weekly step; from February to March, 9 markets spread and 2 concentrated beyond noise.

Between March and June 2026, ChatGPT answers were replaced by ChatGPT Search and Google AI Overviews by Google AI Mode in our tracking. On the ChatGPT side the median market's concentration score rose by 137 points; 17 of 50 markets concentrated beyond noise and 8 spread. On the Google side it fell by 214 points; 30 markets spread and 3 concentrated.

This was the largest step in the year. It mixes a product change, a model change, and a change in how often each prompt was asked, so we report it separately and do not compare across it.

From June to September 2026, 22 of the 50 largest markets spread beyond noise on ChatGPT Search and 12 concentrated. On Google AI Mode, 19 concentrated and 9 spread.

The same market can move in opposite directions on the two engines. In Insurance Quote Comparison Tools, Google AI Mode gave The Zebra 48.6% of #1 picks in September, up from 22.4% in June, while ChatGPT Search spread its picks. In LLM Agent Frameworks and Tooling, LangChain LangGraph rose from 33.3% to 49.3% on Google AI Mode and fell from 47.3% to 37.9% on ChatGPT Search. On ChatGPT Search, RingCentral fell from 50.3% to 26.4% in Business VoIP Phone Systems.

For each of the 50 largest markets we compared the brands named this month with the brands named the month before. From July to September 2026, 34% to 49% of the brands in a market's answers were not named there the month before. Those new brands carried 8.5% to 14.9% of the market's mentions.

Most of that turnover is sampling. When we split one month's answers into two halves, 33% to 39% of the brands in one half are missing from the other. Comparing halves of consecutive months instead gives 35% to 50%. Real month-to-month turnover beyond sampling was 2 to 16 points, largest from June to July. The brands that come and go are the long tail; the brands at the top stay.

A permanent incumbent is a brand that took at least half of a market's #1 picks in at least 6 of the 9 measured months. On the ChatGPT side, 91 of 1,335 measurable markets have one; on the Google side, 87 of 1,338. 26 and 19 of them held it in all nine months, across the product change.

64 market-brand pairs qualify on both engines. DraftKings averaged 88.9% of ChatGPT-side and 94.8% of Google-side #1 picks in Daily Fantasy Sports Platforms. Canva, dbt, Oura Ring, Chainlink, Masterworks, JBL, Google Flights, and All-Clad held the majority in all nine months on both engines.

We counted brands an answer names directly. The #1 pick is the directly named brand that appears first. We measured nine months: November 2025 to March 2026 (ChatGPT and Google AI Overviews) and June to September 2026 (ChatGPT Search and Google AI Mode). We excluded April 2026, when a fault dropped brand detection to almost zero, May 2026, a partial changeover month, and the partial months of October 2025 and October 2026 from all trends. First-named dates use every answer, including those months.

Every trend uses a fixed panel: prompts answered in all nine measured months. Market trends use the 50 largest markets on that panel (18 to 47 prompts each). Concentration scores are corrected for sample size. The noise threshold comes from splitting each prompt's answers within a month into alternating halves. Early brands are defined by the first four weeks of tracking, when fewer prompts were tracked, so the group leans toward often-named brands. This is a measure of how long brands keep AI's #1 pick, not of company age; we do not have founding dates. Some brands exist as more than one record in the index; that splits their counts and makes incumbent shares look smaller, not larger. Figures are an observed cut of Parse data over the stated window.

AI answers did not narrow to fewer brands over the year, but the top slot stayed with the brands AI already knew. For a new brand, being named is common and being named first is rare: 15% of the newest brands reached #1 even once, and most that reached the top three did it on their first day.

Start with specific buyer questions where your brand has a clear advantage. Track progress by engine and market over time, then build on the questions where you begin to appear.