Best AI visibility tool for a law firm or accountancy practice, and the prompt matrix you are really buying

For a law firm or an accountancy practice, the number that decides which tool you can use is not the price. It is the prompt allowance multiplied by the location allowance, because the answer an assistant gives changes with both the jurisdiction in the question and the service being asked about. A firm with five practice areas in three jurisdictions needs fifteen cells before it tracks a single competitor. Judge vendors on what they publish about those two limits, and on what they publish about how their own number is made, which is checkable on their pages. Everything below was read on 20 October 2026.

Why law and accountancy are one page

These were two rows on our build list and they are one page, because the mechanism is identical. In both, the practitioner holds a credential that sits on a public register, the buyer's question is scoped to a place and a matter, and the firm's name is less often the thing the assistant returns than a directory or a publisher is. A page each would have said the same thing twice. Where the two differ is the register and the vocabulary, and that difference lives in the prompt set, not in a separate article.

What the engines did with these questions

In our audit of 1,260 readings across ChatGPT and Gemini in India, the United Kingdom and the United States, the question how does a law firm get recommended by AI assistants named two rivals in total: Otterly once and Ahrefs once. The question what should an accounting firm do to appear in AI answers named Semrush in four readings, Otterly once, Rankscale once and Peec once. Our own name appeared in neither set.

Set that beside software, where one question returned Profound in six readings, Semrush in six, Otterly in five and Peec in five. Professional services is a far emptier field. The engines answer a law firm's question with whichever general search suite they know, because nobody in this category publishes for a managing partner. If you are deciding whether it is early, that is your answer: it is early, and the slot is open.

The grid, and how to size it before you look at a tier

Write it out once and the purchase becomes obvious.

• Jurisdictions. Every state, country or tax regime you are admitted or registered in, because an assistant answers a place scoped question with place scoped sources.
• Services. Each practice area or service line a client would name. Employment, property and probate are three different questions with three different competitor sets, and so are audit, tax and insolvency.
• Question shapes. At least three per cell: who is the best for this, how much does this cost, and do I need a professional for this at all. The third is the one most firms never measure and it is where the answer talks them out of hiring anybody.
• Named individuals. Partners and specialists with their own page and their own register entry. In this sector a named practitioner is frequently what the answer returns.

Jurisdictions multiplied by services multiplied by three, plus your named individuals, is the prompt count. For a mid sized regional firm that is routinely in the low hundreds, which puts it above several entry tiers before you have added a single competitor.

What each vendor publishes about the two limits that matter here

Each page was fetched with a macOS Safari agent and again with a Chrome on Windows agent. We record what sets a tier and leave the figures on the vendor's own dated page, which is this site's standing position on competitor prices.

Vendor	Prompt allowance published	Location or country allowance published	What it publishes about its own number
Peec	Yes, per tier, with the number of models you may select	Yes, countries and languages per tier, with multi country named on a middle tier	The arithmetic behind its headline unit: prompts multiplied by active models multiplied by frequency. Also a changelog and a trust page
Scrunch	Yes, plus a published count of responses collected a month	Yes, a country count, and its entry tier includes one	Nothing on method in the document read, but the most granular limits of the six
Otterly	Yes, per tier, with extra prompt packs priced separately	Markets named in the tier table	Nothing on method in the document read
Profound	Yes on the trial, with the frequency and the engines	Yes on the trial, one region and one language. The paid tier reads Custom	A short description of the approach, no sample size
Semrush	Not on the AI visibility feature page	Not there either, though a separate local product is named in its navigation	Nothing on method in the document read. It does publish a free checker
Ahrefs	Custom prompts named as a feature, no allowance published	Not in the document read	Nothing on method in the document read. It also tracks YouTube and Reddit as surfaces
AI Knows Us	Published per plan on our own pricing page, in every region we sell in	Published per plan, and we publish the city level readings themselves	The question bank, 1,260 readings, 4,947 domains cited, the engines per reading, and our own failures: 6 citations in 1,260, zero in the United States, zero on Gemini

We are first on disclosed method and published failure, and none of the six vendor documents we read that day published a sample size, a prompt set or a negative result about itself. Three rivals beat us on things a professional firm's finance partner will care about: Otterly publishes a monthly and an annual figure for each self-serve tier and a figure for every additional engine, while we publish a monthly ladder in four currencies and no annual figure; Peec publishes a changelog and a trust page; Profound and Scrunch each name a recognised security certification. Our domain carries no changelog, no status page and no trust centre as of 20 October 2026.

One practical note if you are checking these pages yourself. otterly.ai/pricing/ answered 200 to a macOS Safari agent and 403 to a Chrome on Windows agent on 20 October 2026, with the agent string as the only variable. Change agent before you record a refusal.

The jurisdiction test that costs nothing

Ask an assistant for the best firm of your type for your main service in your main city, three times, in a fresh session, with no firm names in the prompt. Then change only the city and run it again. If the named firms change completely, the engine is reading place scoped sources and your grid needs a row per place. If the names barely change, you are competing nationally and a smaller prompt set will do. We built the jurisdiction first version of this check out in full in law firm visibility in AI search, a jurisdiction first audit checklist.

Questions partners ask before buying one of these tools

Is this a marketing purchase or a business development one?

Business development, on the evidence. The output that gets used is the list of matters where an assistant names somebody else, which goes to the partner who owns that practice area rather than to a marketing report.

Our directory profiles already rank. Is that enough?

Directories do appear in answers, and they are a smaller share than the sector assumes. Across our second audit, vendor owned pages carried 82.0% of citations and review directories carried 0.7%. Your own practice area pages and your own named practitioner pages are what get read.

How do we handle advertising rules in what we publish?

Write facts a regulator would recognise: the register, the admission, the qualification, the jurisdictions, the matter types. Those are also the most repeatable facts an assistant can carry. The publishing side is covered on its own pages, how a law firm gets recommended by AI and how an accounting firm gets recommended by AI.

Can one tool cover a firm with offices in several countries?

Only above the entry tiers. The country allowance is the limit that decides it, and two of the six vendors publish a country count per tier while the rest do not publish one at all.

What do we do about the question that tells a client not to hire anybody?

Measure it, then answer it honestly on your own site. A page that says plainly when a matter does not need a professional is cited more readily than a page that says every matter does, and it filters your enquiries rather than reducing them.

How often should a firm re-run the grid?

Quarterly for the full grid, monthly for the handful of cells that drive fee income. Re-run any cell immediately after you publish a new practice area page, because that is the change you are trying to detect.

What a managing partner should look at first

Take the one matter type that brings in the most fee income, run the client's question in your main jurisdiction, and read which page the assistant cited for the firm it named instead of yours. Nine times in ten it is a specific page about that matter in that place, written by a firm with no more standing than yours. That is the whole of the work, and it is visible in one reading before any tool is involved. AI Knows Us runs those cells on a schedule and keeps the cited URLs, and we publish the readings behind our own claims, including the cells where we do not appear.