What ChatGPT and Google AI actually say about R&D tax advisers — our own sweep data
BDO converted every single one of its ChatGPT mentions into a recommendation. On Gemini, in the same sweep, it converted none. The gap between engines is often bigger than the gap between firms.
Steen Stones · Reviewed 5 Aug 2026
Look at any single firm across all three engines in our R&D tax sweep and the spread is often bigger than the difference between one firm and the next. BDO's ChatGPT answers converted every one of its four mentions into a specific recommendation. Its Gemini mentions — five of them, same sweep, same week — converted none.
Avouch multi-engine market sweep, 18 July 2026 — 50 buyer questions, UK R&D tax credit advisers, ChatGPT + Gemini + Claude, 87 firms tallied
The same firm, three engines, three different outcomes
UK R&D tax credit advisers, 18 July 2026 — Gemini shown for comparison, not as a target
Every single one of the 87 firms in this sweep shows a 0% Gemini conversion rate — not a low number, zero, checked against the full tally rather than a sample. ChatGPT and Claude both convert real, substantial shares of their own mentions, and not always for the same firms: ForrestBrown leads on both, but BDO's strength is specific to ChatGPT.
What the recommending answers actually said
Two quotes from the sweep, not yet used elsewhere in this territory, showing the same pattern that holds throughout: a specific reason, not a general endorsement.
“BDO — my pick for larger or technically complex claims, especially where HMRC has opened an enquiry.”
“My pick is Ayming UK for mid-size to large contractors given its specific track record in the sector.”
Where the citations actually come from
forrestbrown.co.uk was the single most-cited domain in the sweep, ahead of kene.partners and empowerrd.com — firms' own sites do appear as sources here, more prominently than in the solicitors sweep, where directories led. That's a real sector difference worth noting rather than smoothing over: R&D tax buyers researching a specific claim appear to read further into a firm's own site than a law firm buyer does, at least in this data.
Common questions
- Is this a different sweep from the pillar page?
- No — this is the same 18 July 2026 sweep, read in more depth. Unlike the solicitors territory, where a second Southampton sweep gave genuinely new data for this page, only one R&D tax sweep exists at the time of writing. We're saying so plainly rather than implying otherwise.
- Why is Gemini's conversion rate zero across every single firm?
- We can see the pattern clearly but not the cause from this data — Gemini appears to default to listing options for this kind of comparative professional-services question rather than picking one, in every sweep we've run so far, including a completely unrelated solicitors market. It's exactly why every number in this territory is reported per engine, never blended.
- Should firms own websites still matter less than third-party sources?
- Our wider research says a firm's own site is never sufficient on its own — it was never once the sole source behind a recommendation across 1,828 re-checked cases in a separate study. What this page shows is that in R&D tax specifically, a firm's own site appears to carry more weight ALONGSIDE other sources than it did in the solicitors sweep — not that it can carry a recommendation alone.
AI can't recommend what it doesn't know.
Start free. No commitment.