The headline numbers
If a buyer asked the five major AI engines for a commercial insurance broker this quarter, three names dominated the answers. Meridian Risk Partners held the top slot for the ninth consecutive month, appearing in 41 percent of all recorded recommendations. But the story of Q2 is churn: of the ten brokers most recommended in March, only seven were still being recommended in May.
That churn rate matters more than the league table. AI engines are not search rankings; they re-evaluate their sources continuously, and a brand that stops earning citations can fall out of the answer in weeks, not years. Two of the brokers who vanished this quarter had strong Google rankings throughout. Rankings did not save them.
The index: top ten by share of recommendations
Share of recommendations is the percentage of recorded answers in which each brand was named, across all four engines, weighted equally. Movement is against the Q1 index.
| # | Brand | Share | vs Q1 | Engines naming |
|---|---|---|---|---|
| 1 | Meridian Risk Partners | 41% | +3 | 5/5 |
| 2 | Halcyon Brokers | 33% | -4 | 5/5 |
| 3 | Crestline Cover | 24% | +24 (new) | 4/5 |
| 4 | Bram & Wexley | 19% | -2 | 4/5 |
| 5 | Saxon Marsh Insurance | 14% | +1 | 3/5 |
| 6 | Northgate Risk | 11% | -6 | 3/5 |
| 7 | Pelham Cover Group | 9% | +2 | 3/5 |
| 8 | Ashdown Broking | 7% | +7 (new) | 2/5 |
| 9 | Ferris & Hale | 6% | -1 | 2/5 |
| 10 | Wrenfield Insurance | 5% | -3 | 2/5 |
The quarter's movers
Crestline Cover is the quarter's story. In March the Vault had never recorded an engine naming them. By the last week of May they appeared in nearly a quarter of all answers, and in 38 percent of ChatGPT's. The climb maps cleanly onto their citation record: a commissioned data study on underinsurance in UK manufacturing was picked up by two trade publications in early April, and the engines began citing those articles within three weeks.
Northgate Risk moved the other way. Their share fell six points after their highest-cited source, a 2023 industry roundup, dropped out of the engines' answers. Nothing replaced it. Their website content has not changed since January; the engines simply found fresher sources for the same questions, and those sources talk about other brokers.
What moved the answers
Three patterns from this quarter's data, consistent with what the Vault sees across categories:
Fresh third-party citations beat old authority. Every riser this quarter earned new coverage in sources the engines already trusted. No riser got there on their own website alone.
Engines disagree, and the gaps are exploitable. Crestline cracked ChatGPT and Perplexity months before Gemini noticed them. A brand watching engine-level data can see which doors are open.
Absence is silent. The two brokers who vanished from ChatGPT's answers saw no alarm, no notification, no traffic graph falling off a cliff. The deals simply started forming elsewhere. This is why we keep saying: if you do not measure the answers, you will not know.
The next UK Insurance Broking index publishes in September. Monthly movers reports publish in the first week of each month.