What we are finding
Research and data from real prompt panels run against ChatGPT, Claude, Gemini, and Perplexity.
One AI engine goes null on agency buyers 8 times out of 10. Another almost never does.
The overall organic null rate across 6,616 real responses is 7.4%, but split by engine the gap is wide: ChatGPT 0.5%, Gemini 2.3%, Perplexity 9.9%, Claude 12.8%. Inside project management software's "for agencies" prompts specifically, Claude goes null 79.6% of the time and Perplexity 35.2%, while Gemini and ChatGPT stay under 1%.
Read the analysisOne qualifier makes AI stop naming any brand you track, a third of the time
Across 6,616 real organic AI responses in three B2B SaaS categories, 7.4% named none of the six brands a panel tracks, not the target and not any competitor. In project management software, the qualifier "for agencies" pushes that rate to 35.2%, because the engine answers with agency-specific tools outside the tracked set entirely.
Read the analysisHalf the time a brand makes the list, it does not become the pick
Across 270 live audit panels, brands named at a specific position in an AI's comparison list convert to the actual recommendation only 49.1% of the time. HR software converts at 66.9%, customer support at just 43.1%. Making the list and being the pick are different outcomes, and the gap between them is where the real buying signal lives.
Read the analysisOn Claude, HR software has almost no second place
In HR software loss records, Claude names BambooHR in 96.7% of competitor mentions and Rippling in just 3.3%. Gemini and Perplexity show a real contest for second place, with Rippling picking up over 30% of mentions on both. Same category, same brand pool, but which engine your buyers use changes who you are actually competing with.
Read the analysisThe same domain was cited in all 173 project management software audits we ran
One domain, thedigitalprojectmanager.com, appeared in the citation list of every one of 173 separate brand audits we ran in project management software, regardless of which brand or competitors the panel was about. The same pattern holds in customer support and HR software. Retrieval is decided by what already ranks for the buyer's query, not by how good your own page is.
Read the analysisWhy a provider error does not corrupt your AI visibility score
Across 271 real audit batches in our own run history, 218 of them (80.4%) hit at least one API error and 269 (99.3%) still completed and got scored anyway. The other 2 halted on purpose rather than finish on bad data. Here is exactly what happens to a failed call between the error and your score.
Read the explainerA head-to-head prompt narrows the field. An open question widens it, without you in it.
Ask an AI engine "Brand X vs Brand Y," and the target brand appears in 90.8% of responses, but the answer rarely goes past the two brands named in the question (avg 2.13 of 6 tracked brands). Ask an open question instead, and the field widens to 3.2 to 3.7 brands, while the target brand's odds of being one of them drop below 1.5%.
Read the analysisThe more specific the question, the smaller your shot
Adding a use-case qualifier to a buyer-intent AI prompt ("...for remote teams," "...for agencies") roughly halves a challenger brand's mention rate: 0.58% with a qualifier versus 1.21% without, across 5,177 real organic recommendation responses. Some qualifiers close the door almost entirely.
Read the analysisA single AI answer rarely names more than three brands
Every audit panel tracks a target brand plus five named competitors, six names total. Across 6,616 real buyer-intent responses in three B2B SaaS categories, the average response named 2.9 of those six, and only 0.3% named all six. The room in a single AI answer is smaller than most brands assume.
Read the analysisYour visibility problem might be a single-engine problem
In customer support software loss records, 171 of 174 competitor mentions (98.3%) come from one engine: ChatGPT. In project management software, the same measurement spreads across all four engines with none above 34%. The fix that works for one pattern does nothing for the other, and the blended score cannot tell you which pattern you are in.
Read the analysisThe competitor you lose to depends on which AI engine you ask
Asana is the top competitor in project management software loss records on every engine we tested. But the number two spot is not the same brand everywhere. Gemini and Perplexity name Jira as the real second option, close to a third of loss mentions. ChatGPT and Claude barely mention Jira at all and name ClickUp instead. A brand checking only one engine will draw the wrong conclusion about who it is actually competing against.
Read the analysisThe same brand, three different prompts, three completely different visibility scores
A recommendation prompt retrieves category ranking pages. A research prompt retrieves explainer content. A comparison prompt retrieves review sites. Each surfaces a different page set, and a different set of brands. In 270 live B2B SaaS panels, 87.4% of brands scored zero on organic recommendation prompts while scoring much higher on branded query types within the same panel.
Read the analysisThe AI visibility ceiling: why even top B2B brands miss 20-56% of buyer queries
We measured 270 B2B software brands across four AI engines. The best-performing brand in customer support achieved an 80% mention rate. In project management, the ceiling was 52%. In HR software, 44%. No brand in any category came close to 100%. The gap is structural, and understanding it changes how you set visibility targets.
Read the analysisWhy a content fix takes 2 to 6 weeks to show up in your AI visibility score
Three separate delays compound between the moment a brand makes a content change and the moment it registers in a re-run prompt panel: the lead time to get the page updated, the search engine re-indexing window, and the time for the page to reach the retrieval threshold the AI engine applies when grounding its answers. Understanding each delay tells you when to re-check and what to expect.
Read the explainerIn B2B software, AI recommends from a pool of 3 to 5 brands per category
Across 270 live AI prompt panels, buyer-intent prompts produced only 12 unique brand recommendations total. Per category: 3 brands in HR software, 5 in customer support, 4 in project management. The AI recommendation list is not a ranking. It is a short list, and most brands are not on it.
Read the analysis87% of brands that appear in AI answers never get recommended
Across 270 live AI prompt panels, 247 brands were mentioned at least once. Only 31 ever received a recommendation. The other 216 appeared in AI answers but never got the buy signal. A look at why mention rate and recommendation rate are almost entirely different metrics.
Read the analysisWho fills the space you leave vacant in AI search
When your brand does not appear in an AI answer, the response does not say nothing. It recommends someone else. Across four categories, 2-4 brands capture 97-100% of the responses that exclude a given brand. A look at who those brands are and why the concentration is this high.
Read the analysisWhat a good AI mention rate actually looks like across 270 B2B software brands
Most B2B software brands cluster between 11% and 30% AI mention rate. The median is 20%. A score of 25% puts you at the 60th percentile. Only 13 of 270 brands exceed 40%. Without knowing the full distribution, any individual score is impossible to interpret. Here is the map.
Read the analysisOne B2B software category has a 4x higher floor for complete AI invisibility
Across 270 live-engine teaser panels in three categories, 11.6% of project management software brands had zero AI mentions vs 2.6% for HR software. The medians are similar, but the floors are not. Category selection predicts visibility difficulty before any content strategy begins.
Read the analysisIn a 46-prompt panel, 11 responses named no internet provider. All 11 came from three prompts about the category in general.
When an AI engine gets a prompt like "What is an internet service provider?", it answers about the category in general and names no specific brand. Every such response lowers your apparent mention rate without telling you anything real about your visibility. A breakdown of what causes null responses, how geographic anchors eliminate them, and what a zero-null panel actually looks like.
Read the analysisAI engines cite whatever the web currently says about your brand
If a directory or aggregator page has outdated information about your brand, an AI engine may report it as current fact. Getting onto the right pages is only half the work. What stale data looks like in practice, why your own site cannot fix it, and how to identify which third-party pages are worth auditing first.
Read the explainerWhen a competitor shows up in AI answers and your brand does not
The cause is almost always a source gap: the competitor is present on a page the engine retrieved when building its answer, and your brand is not on that page. How to identify the specific page from the citation source list, what the pattern looks like across categories, and which platform to target first.
Read the explainerThe typical brand appears in 1 in 5 AI responses. It gets recommended in almost none.
Across 241 B2B software brands in three categories, 10 buyer-intent prompts per brand across four AI engines: 91% had a recommendation rate of zero. The median mention rate was 20%. Complete invisibility is not the common problem. Being mentioned without being chosen is.
Read the analysisChatGPT, Claude, and Perplexity show you their sources. Gemini does not.
In a 46-prompt panel, three AI engines returned citation URLs pointing to specific, readable pages -- comparison aggregators, directories, the brand's own site. Gemini returned 591 citation URLs and 590 of them were opaque redirects through a Google intermediary with no auditable domain underneath. Why this breaks the standard fix playbook for Gemini and what to do instead.
Read the analysisG2 is not driving AI visibility for most software categories. This is what is.
We analyzed citation sources across 270 live brand panels in three software categories. G2 does not crack the top five for two of them. The actual gatekeepers are category-specific editorial sites -- and getting coverage there requires a different playbook than collecting reviews.
Read the analysisPerplexity read this brand's pages 12 times. Nine times it still recommended someone else.
Getting your URL into an AI citation list and getting named in the answer are two different outcomes. A real 46-prompt panel shows the gap: 12 citations, 8 mentions, and 9 cases where the engine retrieved the brand's pages and still picked a competitor. Per-engine data shows how the pattern varies, including Gemini naming the brand 14 times without ever citing its URL directly.
Read the analysisSchema and llms.txt assume the engine reads your site. Often, it does not.
The technical recommendations for AI visibility (schema markup, llms.txt, structured data) improve how engines parse your pages. They do not change which pages engines retrieve. In many categories, the engine never fetches your site at all, because it is answering from directories and forums. The fix has to match where the engine actually looks, not where your website lives.
Read the analysisYour AI visibility check has an expiration date
The four AI engines update their search behavior on independent schedules. A gain on one engine and a loss on another can cancel out in a blended score while the actual picture shifts significantly underneath. Why a one-time audit is a starting point, not a tracking system.
Read the analysisSixty prompts asked AI to recommend an internet provider. This brand appeared in none of them.
Across 60 buyer-intent recommendation queries for one brand, the AI never surfaced the brand once. A breakdown of what buyer-intent queries look like versus branded queries, and what each type returned from a real 46-prompt panel across four engines.
Read the analysisChatGPT cited Facebook zero times. Perplexity cited it 44. That explains the visibility gap.
Same 46 prompts, same week. Perplexity mentioned one regional brand 17 percent of the time while ChatGPT and Claude mentioned it 33 percent. We extracted the citation URLs each engine returned and found Perplexity's single most-cited domain was facebook.com, which ChatGPT never cited once. The per-engine source tables show why the same brand scores differently depending on which AI your buyers use.
Read the analysisA brand appeared in 28 percent of AI responses. The recommendation rate was under 1 percent.
Across 184 AI responses for one brand, 52 included a mention. Only one was an actual recommendation, and it came from a branded comparison query. How to read the gap between mention rate and recommendation rate, and why it matters for new-customer acquisition.
Read the analysisOne AI engine mentioned this brand half as often as the others
Across 46 buyer-intent prompts, one brand's blended mention rate was 28%. Split by engine, ChatGPT and Claude mentioned it 33% of the time and Perplexity just 17%. Why a single AI visibility number hides which engine you are losing.
Read the analysisYour AI visibility number is probably lying to you
An ISP had 28% AI visibility on paper. When we separated branded queries from buyer-intent searches, organic visibility was zero. Why the blended score is the one to stop trusting.
Read the analysisFive URLs are answering your buyers' AI questions
We counted every source cited in 184 AI responses for one local category. The same 5 aggregator URLs appeared dozens of times. A single Facebook group post was cited 23 times. The brand we audited was on none of them.
Read the analysisHow we measure AI search visibility
Our methodology: buyer-intent prompt panels, web-grounded queries across four engines, share-of-voice scoring, and the organic vs branded split most reports skip. Open process, no black box.
Read the methodology