We Measured Whether Organic Traffic Predicts Getting Named by AI, Across Four Categories and Seventeen Brands

ClickUp draws 325,467 organic visits a month. Microsoft draws 145,133,761. When we asked five assistants the same six questions about project management software, both were named in exactly 10 answers.

That is a 446 times difference in traffic producing no difference in citation. It is the sharpest example in a set of seventeen brands we measured across four categories, and the pattern it points at held in every category we looked at: organic traffic is a weak predictor of whether an assistant names you.

Weak, not absent. That distinction is the whole point of publishing the numbers rather than the conclusion, so here they are.

What we measured

For each category we put six buyer questions to five assistants: ChatGPT, Claude, Perplexity, Grok and Google AI Mode. Thirty answers per category. Every answer was stored rather than scored, and a citation was counted when a brand name appeared as written inside an answer.

The brands in each table are not a list we picked. They are the names the assistants themselves produced most often, which matters: choosing the comparison set yourself is how you accidentally prove whatever you set out to prove. Monthly organic traffic is the DataForSEO Labs estimate for each brand’s root domain, United States, pulled the same day as the citation run.

The four categories

The figure in parentheses is Spearman rank correlation between traffic and citations within that category. All four are positive. None is strong, and at four or five brands per category none of them reaches statistical significance. Pooled across all seventeen brands the correlation is +0.29.

The result that survives the small sample

Correlations on seventeen points are suggestive at best. What is not a matter of statistics is the ordering, because you can check it row by row.

In all four categories, a brand with less organic traffic was named in more answers than a brand with more. Not once as a curiosity. Four times out of four.

• Peec AI, at 1,132 visits a month, was named 7 times. SE Ranking, at 320,422, was named 4. That is 283 times less traffic.
• Asana, at 1,877,250 visits, was named 12 times. Microsoft, at 145,133,761, was named 10. That is 77 times less traffic.
• Drata, at 36,607 visits, was named 14 times. Hyperproof, at 61,813, was named 7.
• Shopify, at 3,452,110 visits, was named 9 times. Salesforce, at 8,436,893, was named 8.

The efficiency spread is the other durable number. Citations per hundred thousand visits runs from 0.01 for Microsoft to 618.37 for Peec AI. Two brands can sit in the same answer with four orders of magnitude between what it cost them in traffic to get there.

What we think is going on

A ranked results page is a competition for position among pages. An assistant answer is a competition for a mention among entities, and the model is not reading a traffic estimate when it composes one. It is drawing on how often, how clearly and how recently a brand is described as the thing being asked about.

That favours specificity over size. Vanta and Drata are compliance automation and almost nothing else, so text about compliance automation names them. Microsoft is an enormous number of things, and its project management product is a fraction of what the internet says about it. Traffic measures the whole domain. Citation measures the association.

If that reading is right, the practical consequence is more encouraging than the usual SEO arithmetic: you are not required to out-rank a giant to get named beside it. You are required to be more clearly the thing.

What this does not show

Four categories and seventeen brands. That is enough to demonstrate the ordering happens and not enough to say how often it happens in general.

Citation counts come from one run per category, so they are a snapshot. Assistants vary between runs, and we did not average across days. A brand sitting one mention apart from another in these tables should be read as a tie.

Counting a brand as cited when its name appears as written is a deliberately strict rule, and it undercounts. When an assistant writes “Zendesk Suite” the rule does not credit Zendesk. We prefer undercounting to a fuzzy match that quietly inflates, but it is a choice worth knowing about when you read the numbers.

We also did not test causation in either direction. Nothing here says publishing more pages produces more citations. It says the brands getting cited are not reliably the ones with the most traffic.

The raw data

Every number above is in citations-vs-traffic-2026-08-25.json, including the per-category correlations and the method note. If you re-run it and get something different, we would rather know.

We built FoundCite because we wanted this measurement for our own category and could not buy it. It runs the same six-questions-across-five-assistants process for any domain, stores every answer, and reports counts rather than a score. If you want to see where your own brand lands on a table like these, the free scan needs no account. If you want the longer version of why counts beat scores, that is in what a GEO score actually measures.