AI Visibility Benchmarks by Category: What 3,011 B2B Domains Reveal

An AI visibility benchmark is a measured baseline for how a category performs on two separate axes: how much organic traffic its companies have lost, and how often AI answer engines name those companies when a buyer asks a category question. The two move independently, which is why a single "AI visibility score" hides more than it reveals.

Most AI visibility benchmarks you will find are vendor scores with no published denominator. We built ours the other way round. peppereffect scanned 3,205 B2B domains for a specific signature, organic clicks collapsing while impressions held, verified 3,011 of them, and then asked a live AI engine the real buyer question for each company's category. This article publishes what came back, category by category, along with the parts of the method that limit what the numbers can prove.

What you will find below:

• The full 17-category benchmark table, with an interactive filter so you can isolate your own category
• Why the category with the worst traffic decline also has one of the best citation rates, and what that tells you
• The exact measurement method, including the three things it cannot prove
• How the published third-party studies compare, and which widely repeated numbers do not survive a source check
• A five-step process to run the same measurement on your own domain

What is an AI visibility benchmark, and why do category averages mislead?

An AI visibility benchmark is a reference point, not a verdict. It tells you how companies structurally similar to yours are performing on a measurable axis, so that your own number means something. Without a denominator, a visibility score is a number with no scale attached. "Your AI visibility is 34" is unusable unless you know what 34 means against companies in your category.

The catch is that most benchmarks in this space are built from whatever set of brands a vendor happened to track, with no stated selection rule. Ours has a selection rule, and it is aggressive: every domain in this dataset had already lost at least 45% of its peak organic traffic before it entered the sample. This is a cohort of companies that have already been hit. It is not a random sample of B2B, and none of the numbers below should be read as "B2B lost 64% of its traffic."

That distinction matters more than it sounds. If we told you 87% of B2B domains have lost half their organic traffic, that would be false. What is true is that among domains already showing the traffic collapse signature, 86.8% have lost at least half, and 27.7% have lost three quarters or more. The benchmark answers "how bad does it get once it starts," which is the question a CMO watching impressions hold while clicks fall actually needs answered.

How much organic traffic did the collapse cohort actually lose?

The median domain in the cohort lost 64% of its peak organic traffic. The middle half sits between 54% and 76%. At the ninth decile the decline reaches 85%, and 5.4% of domains lost 90% or more of what they once had.

Aggregate the whole cohort and the picture sharpens. These 3,011 domains once drew a combined 734.0 million monthly organic visits. The latest measurement puts them at 244.9 million. The cohort retained 33.4% of its peak organic traffic, which means two thirds of it is gone.

Source: peppereffect collapse scan, 3,011 verified domains, measured August 2026. Traffic figures are third-party estimates of peak versus latest monthly organic visits.

The scan itself began with 3,205 candidate domains. We discarded 65 as false positives, where the apparent decline turned out to be a measurement artifact such as a domain migration, and flagged 129 more for manual review. Only the 3,011 that survived verification are counted here. Publishing the discard rate is deliberate: a benchmark that never rejects any of its own inputs has not been checked.

Worth noting what this cohort is not. Every domain here was selected because it had already declined, so the dataset cannot tell you what share of all B2B companies are affected. It can tell you, with reasonable confidence, how the damage distributes across categories once a company is in trouble, and that distribution turns out to be far from uniform.

Which B2B categories lost the most organic traffic?

Customer success technology is the worst hit category in the dataset, with a median decline of 69%. Compliance and risk fared best at 60%. That nine point spread looks modest until you translate it: at the 75% severity threshold, customer success (33.8%) has more than one and a half times the proportion of severely damaged companies that proptech does (22.4%).

Use the filter to isolate your own category. The retained column is the one most people miss, and it is the more honest measure of category-level damage, because it weights by traffic volume rather than treating a 500-visit site and a 500,000-visit site as equals.

Source: peppereffect collapse scan, 3,011 verified B2B domains across 17 categories, measured August 2026. Median decline is peak versus latest monthly organic visits per domain. Traffic retained is category-wide latest visits as a share of category-wide peak, so it weights larger sites more heavily.

Compliance and risk is the instructive row. It has the mildest median decline in the table at 60%, and the second-worst traffic retention at 26.3%. Both are true. The typical compliance company lost less than the typical customer success company, but the large compliance sites, the ones carrying most of the category's traffic, were hit hard enough to drag the weighted figure down. If you benchmark yourself on the median alone you will conclude your category got off lightly, and if you are one of the larger sites in it you will be wrong.

How often do AI answers actually name your company?

The second measurement is separate from the first. For 926 of these companies we took the genuine buyer question for their category, the kind a prospect types when they are choosing rather than researching, and put it to a live AI engine with web search enabled. Then we recorded whether the company itself appeared in the answer, and which vendors appeared instead.

The company was named in 234 of 926 answers. That is 25.3%. In the other 74.7%, a buyer asking the defining question of that company's own category got an answer that did not include it.

Source: peppereffect category question runs, 926 companies with a definitive result, August 2026. One question per company, put to GPT-4o mini with live web search at temperature 0. Categories with fewer than 25 companies tested (insurtech 19, customer success 20, construction tech 22) carry wide error margins and should be read as directional only.

Proptech at 6.1% is the number that should worry that industry. Two companies out of 33 were named when a buyer asked the question their whole category is built around. This is the layer that Share of Model exists to measure, and it is invisible in Google Analytics, because nothing happened. No impression, no session, no bounce. The company was simply absent from the conversation.

Does the answer set concentrate around a few big names?

No, and this is the most useful finding in the dataset for a mid-market company. Across the answers we recorded, 1,914 distinct vendors were named a total of 2,577 times. 81.6% of those vendors appeared exactly once. The single most frequently named vendor showed up 15 times, which is 0.6% of all mentions. The ten most-named vendors together accounted for 3.6%.

That is a wide open field. Compare it with Semrush's 2026 AI Visibility Index, built on 126 million US prompts across 22 industries, which found the top three brands controlling 82.9% of visibility in news and media and 76.9% in consumer electronics. Concentration like that is real, but it is a property of large consumer categories with famous brands. In the same study, finance came in at 41.4% and industrial at 42.2%, and our mid-market B2B software categories are more dispersed still.

The practical reading: in consumer electronics you are fighting three incumbents for a slot. In mid-market B2B software, no one owns the answer yet. The model is assembling a shortlist from a very long tail, which means the slot is winnable in a way it is not for a company trying to displace a household name. That is a narrow window, and it is the entire argument for treating answer engine optimization as urgent rather than experimental.

Why do traffic loss and citation rate move independently?

Look at marketing technology. It has the second-worst median decline in the dataset at 67%, and the highest citation rate at 44.7%. Customer success has the worst decline at 69% and one of the lowest citation rates at 10.0%. If citation protected traffic, those two rows would not look like that.

The likeliest explanation is that martech companies publish relentlessly and are written about constantly, which builds exactly the signal that gets a brand named. Ahrefs' study of 75,000 brands and AI Overview mentions found branded web mentions the strongest correlate at 0.664, roughly three times the correlation of backlinks at 0.218. Martech is a category that talks about itself in public more than almost any other. It earns the mention. It still lost the click, because the mention happens inside an answer the buyer never leaves.

The same pattern shows up in an independent dataset. SE Ranking tracked B2B review platforms from January 2024 to December 2025 and found TrustRadius down 92.2%, Capterra down 89% and G2 down 84.5% in US organic traffic, while those same platforms remained among the most heavily cited sources in AI Overviews. Maximum citation, collapsed traffic. Being the source the model quotes is not the same as being the destination the buyer visits.

What do the published studies actually support?

Our data covers one cohort and one engine. The published research covers different ground, and it is worth being precise about what each study does and does not establish, because this field repeats numbers carelessly.

Sources: SparkToro, Ahrefs CTR study, Pew Research Center, Orbit Media, Ahrefs brand correlation study.

Two things stand out. First, the Orbit Media study is the one to keep. AI sources produced just 0.5% of all sessions across those 97 B2B sites, and converted at nearly four times the rate of organic search on high-intent actions. That is the answer to the objection that AI traffic is too small to matter: it is small and it is disproportionately the traffic that turns into pipeline.

Second, be careful with the numbers circulating in this space. While writing this article we cut a figure from our own internal reference, a much-repeated claim that AI answers reduce position-one clickthrough by around 58%. The primary Ahrefs source reports 34.5%, and we could not trace the higher figure to any primary study. If you are citing a statistic in a board deck, open the original and read the methodology section. A surprising number of the numbers in this field do not survive that.

The pattern to notice across all five studies is that none of them measure the same thing, and none of them measures what you actually need to know, which is whether your company gets named for your category's buying question. That number is not in any public dataset. It has to be measured for your domain specifically, which is why getting cited by ChatGPT starts with measurement rather than tactics.

How do you run this measurement on your own domain?

You can reproduce the method without our dataset. It takes an afternoon and it is more informative than any vendor score, because you choose the questions.

Frequently Asked Questions

What is a good AI visibility score?

There is no universal number, which is why category benchmarks matter more than scores. In our dataset the average company was named in 25.3% of answers to its own category's buying question, but the range runs from 6.1% in proptech to 44.7% in marketing technology. A 30% citation rate would be strong in proptech and below average in martech. Judge yourself against your category row, not against an absolute threshold, and treat any tool that gives you a single context-free score out of 100 with suspicion.

How do you measure AI visibility?

You define the questions your buyers ask, run them across the major engines on a fixed schedule, and count how often you are named and who is named instead. The measurement has two layers that are easy to conflate: whether the model mentions your brand in its text, and whether it cites your domain as a source. Those need separate tracking, because they have different causes and different fixes. Our five-step method above is reproducible by hand, and the Share of Model metric is the formal version of the same idea.

Can I run an AI visibility audit myself?

Yes, and for a first read you should. Twenty manual queries across four engines will tell you more than most automated dashboards, because you control the questions and you see the full answer text rather than an extracted score. The limits appear when you want a trend rather than a snapshot, or when you need to compare yourself against a category. Non-determinism means a single run can mislead in either direction, so repeat on a schedule before drawing conclusions. Our guide to optimizing for AI search covers what to do with the findings.

Which AI visibility tools measure this properly?

Most commercial tools track brand mentions across engines and report a score. The questions to ask a vendor are: which engines, how often, whether they let you supply your own prompts, and whether they distinguish mention from citation. The last two matter most. A tool that runs its own generic prompts is measuring a category you may not be in, and a tool that merges mention and citation into one number cannot tell you which problem you have. Ask for the denominator, and if the vendor cannot produce it, the score is decorative.

Does a traffic decline mean AI answers caused it?

Not on its own, and this is where most analyses go wrong. Organic declines of this size have several plausible causes, including Google core updates, site migrations, deliberate content pruning and ordinary competitive loss. What makes the AI hypothesis worth investigating is the specific signature of impressions holding steady while clicks fall, because that shape means you are still being surfaced and no longer being visited. If your impressions fell alongside your clicks, you have a ranking problem, and the diagnosis is different.

Is AI traffic too small to be worth optimizing for?

It is small and that is the wrong frame. Orbit Media measured AI sources at 0.5% of all sessions across 97 B2B sites, which sounds dismissible until you see that those visitors converted on high-intent actions at 1.91% against 0.50% for organic search. More importantly, the citation happens whether or not anyone clicks. When a buyer asks which vendors to consider and receives three names, the competitive damage is done inside the answer. That is a share question, not a traffic question, and volume is the wrong way to size it.

Resources

• SparkToro, In 2026 Less than One Third of Google Searches Still Send a Click
• Ahrefs, AI Overviews Reduce Clicks by 34.5%
• Pew Research Center, Do people click on links in Google AI summaries?
• Orbit Media, AI Traffic Conversion Rates across 97 B2B Websites
• SE Ranking, Review Platforms in AI Overviews
• Ahrefs, An Analysis of AI Overview Brand Visibility Factors
• Semrush, 2026 AI Visibility Index across 126 Million Prompts