Our Pilot AI Visibility Audit of 29 B2B Companies

This is our pilot audit. We have since run 12 larger single-industry audits covering 217 businesses and 1,085 live buyer queries, all on our AI visibility research page. Use those for industry figures. In particular, see the audits of accounting firms, consulting firms, marketing agencies, insurance brokerages and B2B SaaS companies. Those supersede the per-sector averages below. This post remains the overview and the best explanation of the method.

We didn't estimate this. We ran it.

Most "AI visibility statistics" articles right now are built the same way: survey a few hundred marketers about their impression of how visible their brand is to AI, then publish the averages. That measures confidence, not visibility.

We did something more boring and more useful. We took the exact audit tool this site offers visitors for free, the same classify-then-grounded-search pipeline, the same prompts, the same model, and ran it against 29 real companies across six B2B industries we actually work in: accounting, insurance brokerage, management consulting, marketing agencies, recruiting and staffing, and B2B SaaS. No survey. No self-reported confidence score. Each company either got named by a live AI search for a real buyer question, or it didn't.

How the audit actually works: query fan-out

Here's the mechanic underneath every number in this piece, because it matters for understanding what "visibility" means here.

When someone asks ChatGPT, Perplexity, or a Google AI Overview a buying question, the system does not run one search. It breaks the question into several related sub-queries, searches each one, then synthesizes a single answer from whatever came back. This is called query fan-out, and it's the actual mechanism behind every AI-generated answer to a buying question, not a metaphor for it.

Our audit tool does the same thing deliberately, as the measurement method rather than just the thing being measured. For each company, it first reads the homepage and classifies what the business actually sells, in the words a customer would use, not the words the company uses about itself. Then it fans that out into five distinct buyer-intent queries:

• Local (1 query): includes the business's detected location, the way someone searches when they want a nearby provider.
• Category (2 queries): no location, phrased as someone looking for the best or most reliable providers, written to be meaningfully different from each other.
• Problem (1 query): phrased as the underlying problem rather than the solution, the way someone searches before they know what kind of company they need.
• Comparison (1 query): asking who the top providers are, or asking for a direct comparison.

Every one of those five runs as a real, live, grounded AI search, not a guess from the model's training data. The target company's own name is never in the query. The entire test is whether they surface without being named directly, the same way a real prospect would ask.

That gap between comparison queries (69%) and problem-first queries (21%) is the clearest, most actionable pattern in the whole dataset. Most companies' content is written to win the comparison, "best accounting firm for X," because that's the obvious bottom-of-funnel keyword. Almost nobody writes content that answers the problem before the reader knows what kind of company solves it, which is exactly the query type AI assistants currently struggle hardest to answer with anyone's name at all.

The headline number, broken down by industry

42.8% average visibility hides a split wide enough to change how you'd prioritize this work entirely, depending on what you sell.

Marketing agencies and B2B SaaS companies cluster at the top, which makes a kind of obvious sense once you see it: these are businesses whose entire job is being found and chosen online, so the same muscle that wins them clients also wins them AI citations. TopRank Marketing and Close both scored a perfect 100%. Ironpaper, Directive, SmartBug Media and Lattice all scored 60% or higher.

The bottom of the list is where this gets genuinely useful. Management consulting averaged just 17%, and two of the six firms we audited, Cadmus and ZS, a management and marketing science consultancy with real scale and a real reputation, scored a flat 0%. Not "ranked low." Named zero times across five separate buyer queries each.

Every company we audited, industry by industry

Averages hide individuals. Here is every one of the 29 companies, with the exact score, which of the five query types actually surfaced them, and who the AI named instead when they didn't appear.

B2B SaaS

This sector had only two companies in this audit, so treat its numbers as a hint, not a result. For a proper SaaS sample, see our audit of 30 B2B SaaS companies.

Marketing Agency

This sector had six companies in this audit. For a larger sample, see our audit of 19 established marketing agencies.

Worth pausing on Directive and Ironpaper specifically: each named the other as their own top rival. Two direct competitors, both investing seriously in this exact discipline, both showing up for each other's buyer queries. That's what it looks like when a whole category does the work.

Accounting

Recruiting and Staffing

Insurance Brokerage

Management Consulting

This sector had six firms in this audit. For a larger sample, see our audit of 19 established consulting firms.

Why the bottom of the list looks the way it does

This isn't really a story about those specific firms doing anything wrong. It's a story about what happens when an AI system has nothing distinctive to go on.

Across every query where a management consulting or accounting firm failed to get named, the same small set of giant incumbents filled the space instead, repeatedly, regardless of what the actual query was asking.

Seven names account for a disproportionate share of every "who should I call" answer across two entire industries. That is not because a boutique or mid-market firm is a worse answer. Often it is a better, more specific one for the query actually asked. It happens because an AI system defaults to brand recognition whenever a query doesn't give it anything sharper to work with. That is a structural opening, not a wall. It means the firms investing in genuinely specific, query-shaped content have real room to take share from household names that are, in this one narrow sense, coasting.

What it actually looks like to score zero

Numbers like "0%" are easy to skim past. Here is what actually happened, in the AI's own words, for the two firms that scored it.

Cadmus, a real government and technology advisory firm, was asked five buyer-shaped questions about exactly what it does. Every single search came back the same way:

"Searches for 'government technology consulting firms near me' and related ranking queries returned directories, roundups, and named firms such as Booz Allen Hamilton, Accenture, Deloitte..."

"Search results returned broad consulting rankings and roundups... featuring firms like ICF, Tetra Tech, Deloitte, McKinsey, WSP, AECOM, and Brattle Group..."

Cadmus's name never appeared once, not even in a long list. Every query, including the ones phrased around the exact problems Cadmus solves, resolved to the same five or six familiar giants.

ZS, a management and marketing science consultancy with real scale, scored identically. The category query is the clearest example:

"The search results consistently list McKinsey, BCG, Bain, Deloitte, PwC, Accenture, and similar large generalist consulting firms as the top management consulting firms, with no mention of ZS."

That phrase, "with no mention," is the model stating plainly what it found. Not a low ranking. Not a weak mention buried in paragraph four of a roundup. Absent. ZS scored 0% on this run (a later, larger consulting audit scored it 40%, so single scores are snapshots). Even so, for a firm with real scale and a real specialty, it illustrates that reputation among humans does not automatically transfer to visibility inside an AI-generated answer. The AI is not reading a Rolodex of respected firms. It is reading whatever text got indexed and summarizable, and a firm that does not show up in enough specific, query-shaped content simply is not there to be cited.

What it actually looks like to score 100%

The contrast is just as instructive. Close, a sales CRM, and TopRank Marketing, a B2B content agency, were the only two companies in the whole audit named on all five queries.

Close's comparison query:

"Close shows up consistently across multiple CRM comparison roundups and listicles, typically positioned as the top choice for high-volume, phone-heavy inside sales teams..."

Notice the specificity: not "a good CRM," but "the top choice for high-volume, phone-heavy inside sales teams." That's a company that has clearly earned a specific positioning in the content the AI is drawing from, not just general brand awareness.

TopRank Marketing's problem-first query, the hardest category across the whole dataset (21% average appearance rate), still surfaced them directly:

"TopRank Marketing's own B2B SEO/AEO service page appeared directly in search results for this query, describing their work on brand visibility and search rankings for clients like..."

That is a rare outcome in this dataset: a company's own page, not a third-party roundup, answering a problem-first query directly. Only about one in five problem-first queries named a company at all, and own-page citations were rarer still. It is also the most valuable outcome, because it means the content itself did the work, with no dependency on being included in someone else's list.

The citation landscape is far more open than it looks

Here's a finding we didn't expect going in, and it changes the read on everything above.

It would be easy to assume a handful of big aggregator sites control what AI assistants cite, the way a handful of consultancies control who gets named in that industry. The data doesn't support that. Wikipedia and Yelp were the only two sources that showed up repeatedly across genuinely different industries, and even added together they accounted for under 9% of all citations. The other 91% is spread across 710 other domains: trade publications, directories, comparison sites, and, notably, the audited companies' own competitors' websites.

That's the actionable read. There is no gatekeeper to get past. The citation pool is this fragmented because most companies in most categories have not published anything specific enough to earn a citation yet. The opening is real, and it's wider than the "two platforms own everything" story would suggest.

What to actually do with this, depending on your situation

Four findings came out of this audit. Each one points at a different, specific fix, not a generic "publish more content" prescription.

If you're in a crowded, incumbent-heavy category (consulting, accounting, insurance, law). The brand-recognition default is the thing working against you, and generic "about us" or "our services" content will never beat it. What does beat it: content built around the specific, narrow situation a buyer in your actual niche is in, specific enough that a giant generalist firm's content never gets written to compete with it. APCO, the one management consulting firm in our audit that cleared 40%, is positioned narrowly around public affairs and government relations, not generalist strategy. Narrow and specific outperformed broad and prestigious, repeatedly, across this entire dataset.

If problem-first queries are your weak spot (almost everyone's, at 21% average). Audit your own content for whether anything answers a buyer's problem before they know your category exists. Most B2B content is written backward from the solution: "why you need a CRM," "the benefits of outsourced IT." The queries that actually got answered started from the problem itself: "we're losing deals because our team doesn't follow up fast enough," not "what CRM should I buy."

If you've never checked whether your site is even readable. Five of the companies on our original list, Wipfli, US Insurance Brokers, NFP, Alvarez & Marsal, and Athens Group, could not be scanned at all. Their sites returned errors or blocked the request outright. A site that cannot be crawled cannot be cited, regardless of how good the underlying business is. This is a five-minute check, not a strategic project, and it's worth doing before anything else on this list.

If you assumed a handful of big platforms control AI citations. They don't, not even close. 91% of the citations in this audit went to one of 710 different domains, most appearing only once or twice. That means the bar for becoming one of the sources an AI system cites is lower than most companies assume, because almost nobody in most categories has published the kind of specific, query-shaped content that earns a citation in the first place.

Check your own number

Every company above got run through the exact tool below. It takes about ninety seconds and the methodology is identical to everything in this piece: real queries, live grounded search, your own name never in the query.

Do AI assistants recommend you?

Five buyer-style queries for your category, run through live AI search, show who gets named instead of you. Takes about ninety seconds.

Your website address

Run the free audit

Free. The full report arrives by email so the tab can close.

If you'd rather not do this yourself

Everything in the section above works whether or not you ever talk to us. That's deliberate. But it's worth being direct about where Deligatr fits. AEO and GEO are the core of our Inbound package, so weigh this section accordingly.

If you're in one of the crowded, incumbent-heavy categories from this audit, consulting, accounting, insurance, or anything else where a handful of household names soak up every generic query, the fix genuinely is a project, not an afternoon. It means auditing what your content currently answers, finding the specific buyer situations nobody else is writing for, and rebuilding around queries that actually convert instead of queries that make you feel represented. Our AEO strategy work is built around exactly that: getting cited inside AI-generated answers, not just ranked on a results page nobody clicks through anymore.

If you want a second, deeper pass beyond the automated version above, specifically on your own market and your own named competitors, we'll walk through what we'd prioritize first, using your real numbers instead of a generic framework.

Methodology and honest limitations

In the interest of not doing the thing we're criticizing elsewhere in this piece, here is exactly what this audit did and did not measure.

What we did. Fetched each company's homepage, classified what they actually sell in buyer language, fanned that out into 5 distinct buyer-intent queries per company using the exact prompts and model our own live audit tool uses, ran each query as a real, live, grounded AI search, and recorded whether the company was named.

What this doesn't cover. Only the homepage was read per company, not inner service or location pages, so a firm whose homepage is thin but whose service pages are strong may be undercounted here. Five queries per company is enough to fan out across the real buyer-intent categories that matter, but real buyer language is broader than five phrasings. This is one audit run on one day in October 2026; AI search results shift over time, and a rerun next quarter would not produce identical numbers, though we'd expect the industry-level pattern to hold.

On the sample. 29 companies, not a market-wide census. We selected real, identifiable, mid-to-large firms across six industries we work in, weighted toward companies with an actual public web presence worth auditing. A handful of additional companies (Wipfli, US Insurance Brokers, NFP, Alvarez & Marsal, Athens Group) could not be scanned at all, their sites returned errors or blocked the request outright, which is itself a small data point: a site an AI crawler cannot read is a site that cannot be cited, regardless of how good the underlying business is.

FAQ

Related reading

• How to get cited in ChatGPT, Perplexity and AI Overviews
• What is answer engine optimization (AEO)? The complete guide
• AI visibility research: every audit in one place
• AEO vs GEO vs SEO: what actually separates them
• We audited 30 B2B SaaS companies for AI visibility
• Inbound marketing: what actually brings in buyers
• B2B buying signals: the complete guide
• Content marketing strategy: a framework that actually ships