We have now run four categories through ChatGPT, Claude, Perplexity and Gemini: AI SEO tools, email marketing for SaaS, privacy-friendly web analytics, and small business CRM. Five buyer questions each, four engines, twenty answers per category. Eighty answers.
We expected the engines to mostly agree, with noise at the edges. They do not. They split, in the same direction, every time.
The split
| Category | Named only by Perplexity | Named by everyone except Perplexity | Named by all four |
|---|---|---|---|
| AI SEO tools | 4 | 8 | 1 |
| Email marketing for SaaS | 4 | 7 | 3 |
| Privacy-friendly web analytics | 2 | 6 | 3 |
| Small business CRM | 1 | 6 | 5 |
| Total | 11 | 27 | 12 |
Twenty-seven products are recommended by ChatGPT, Claude and Gemini and never once by Perplexity. Eleven are the exact mirror image. Only twelve products in four categories are known to all four engines.
The names make the pattern obvious.
Named only by Perplexity: Semrush, Ahrefs, Frase, Writesonic, ActiveCampaign, MailerLite, Brevo, Encharge, Amplitude, Fathom, Streak.
Named by everyone except Perplexity: GrandRanker, Byword, Contentpen, Outpacer, SEObolt, ZRank, Serpa, Allable, Mailazy, Sable, Grivo, Firstloop, Shipletter, GhostlyX, Glyphex, Tiny Metrics, Ripples, SimpleData, Peek, Sealmetrics, Followupzu, ClaretCRM, Conduyt, Sambandh, Moon CRM, Nutshell, and others.
The first list is companies you have heard of. The second list is mostly companies you have not.
Why this happens
Perplexity's Sonar searches the live web on every question. The other three had web search enabled for these scans too, but they lean much harder on what the model already absorbed during training and fine-tuning.
So the two groups are answering slightly different questions.
Perplexity answers something close to what does the web currently say about this category. That surfaces companies with the most reviews, comparisons, listicles and coverage, which is why it names Ahrefs and Semrush.
ChatGPT, Claude and Gemini answer something closer to what do I associate with this category. That surfaces whatever has been written in a form the model took in, which turns out to favour newer tools that publish heavily and structure their pages for machines.
Neither is wrong. They are different retrieval strategies, and they produce different shortlists for the same buyer.
The part that surprised us
The split is not the same size in every category. It tracks how old the category is.
Small business CRM is a mature category, and five products are named by all four engines: HubSpot, Pipedrive, Zoho CRM, Less Annoying CRM and OnePageCRM. Only one product, Streak, is Perplexity-only.
AI SEO tools is a new category, and exactly one product is named by all four: SEOmatic. Eight are named by everyone except Perplexity.
The older the category, the more the engines converge. The newer the category, the more they disagree, and the more the answer depends on which assistant your buyer happens to open.
If you sell into a young category, there is no consensus answer yet. That is the opportunity and the risk in the same sentence.
What the engines cited
Across all four scans the pattern in the citations is consistent: products in the "everyone except Perplexity" group are usually cited from their own domains.
seomatic.ai, mailazy.com, sequenzy.com, sablemail.com, grivo.io, ghostlyx.com, tinymetrics.io, ripples.sh, followupzu.com, claretcrm.com, conduyt.com, sambandh.io, mooncrm.io all appear in the cited-sources lists of their categories.
These are not household names being remembered from training. They are companies whose own pages were read while the answer was being written. Several of them say plainly that this is deliberate. GrandRanker's homepage advertises content "optimized for AI search (Google AI Overviews, ChatGPT, Perplexity, Claude, Gemini)".
It is working, on three engines out of four.
What to do with this
The useful question is not whether you are visible in AI answers. It is which half you are visible in, because the two failures have different fixes.
Only Perplexity names you. You have live coverage and little presence in the models. Reviews, comparisons and listings are doing their job; the models have not absorbed you. This is the position most established companies are in, and most of them do not know it.
Everyone except Perplexity names you. You have presence and thin live coverage. This is a real position, and it is more fragile than it looks: a model update can remove you from all three at once, and you have no independent search-grounded footprint to fall back on.
All four name you. Twelve products in four categories manage this. It is worth studying what they have in common.
Nobody names you. Start with the first list, because that one is fixable with work you can do this quarter.
Method, honestly
Four categories, five buyer-intent prompts each, asked once per engine to ChatGPT (gpt-4o-mini), Claude (Haiku 4.5), Perplexity (Sonar) and Gemini (2.5 Flash) with web search enabled. Three categories were scanned on 30 July 2026, the CRM category on 18 August 2026.
Counts are of answers, so a product named twice in one answer counts once. Product names are merged across obvious variants such as HubSpot and HubSpot CRM, or Plausible and Plausible Analytics. Parsing is automated and could miss a stray mention.
These are single-day snapshots and rankings move. The scan data behind every number here is committed alongside this post, so any figure can be checked.
Whoaisays tracks whether AI assistants recommend your brand, every week. Free for 14 days, no card required; plans from $29/mo.