AI search optimisation for accountants: what actually gets a firm recommended

If you take one thing from this page: which AI engine you're being measured on matters more than almost anything else about your firm. In Avouch's August 2026 sweep of UK accountants, 50 buyer questions put to ChatGPT, Gemini and Claude with live web search, Gemini named a firm somewhere in its answer 184 times across its 17 responses. It recommended one of them, as its specific pick, zero times. ChatGPT and Claude, asked the identical questions in the identical week, each named a favourite in most of their answers.

We've seen the same shape in our sweeps of solicitors and R&D tax advisers too, Gemini naming firms freely without picking one, in a prompt that explicitly asked it to. Whatever a dashboard reporting blended "AI visibility" across all engines shows you, it is hiding a number worth checking for your own firm: which engine you're actually being measured on, and whether that engine picks favourites at all.

Named is not chosen

Named is any answer that mentions a firm at all, one line among a list of five options. Recommended is narrower: the engine puts that specific firm forward as its pick, the one it tells the buyer to go with. Across our 50-question sweep, a handful of firms did that conversion well. Most firms named did it rarely or never.

Crunch converted 63% of its mentions into a recommendation and Taxfix 58%, while The Accountancy Partnership, named almost as often as Taxfix, converted 20%. Several of the UK's largest, most recognised accountancy networks were named repeatedly across the sweep and picked as the answer not once. Nothing about the questions changed between them. What changed was the evidence each firm gave the engine to point at.

What the engines actually cite as the reason

Because Avouch stores full answer text, we can read the sentence that does the recommending. Across the sweep, the pattern is specificity: a named niche, a stated fee structure, or a review count, never a general reputation claim.

That sentence carries two checkable reasons, a stated specialism and a review-volume claim, and nothing that couldn't be verified by clicking through. Blick Rothenberg converted every single mention it received into a recommendation across all three engines in this sweep, on the strength of one line: "my recommendation for wealthy individuals, inheritance tax, property or international affairs", a specific practice claim, not a superiority claim.

Reviews do more work here than in any territory we've swept

uk.trustpilot.com was the single most-cited domain across the whole sweep, ahead of crunch.co.uk and coxhinkins.co.uk, which tied for second. That is a real difference from the solicitors and R&D tax territories, where a directory (Legal 500, Chambers) or a firm's own site led the citation list. For accountants specifically, the engines appear to lean on third-party review volume as their main proof of trust, more than in any other territory we've measured so far, worth knowing before assuming the same playbook that worked for a law firm transfers unchanged.

What this means for a UK accountancy firm

• Build and maintain a real, current Trustpilot presence. It's the single most-cited source in this territory, ahead of every directory and most firms' own sites.
• State a specific practice claim an engine can quote directly, "we work exclusively with property developers", not "we're a trusted local firm". Every recommending quote in our sweep carried one.
• Check Gemini separately from ChatGPT and Claude. Blending them into one "AI visibility" score hides a real, repeatable, three-territories-running pattern: one engine names firms freely and picks almost none of them.
• Don't assume the solicitors playbook (directory rankings first) transfers unchanged, reviews carried more of the weight here than directories did.

How this was measured

50 buyer-intent questions covering UK accountants generally, sole traders, limited companies, VAT, payroll, specific sectors like property and hospitality, put to ChatGPT, Gemini and Claude with live web search on 5 August 2026, split roughly evenly across the three engines. 187 firms surfaced across 52 distinct source domains. Every answer was read and normalised into named versus recommended, and the underlying verdicts are stored rather than summarised, so any figure here can be traced back to the answer that produced it. Per-engine results are held separately and never blended into an average, for the reason set out above.