AI engines name a specific brand far less often than most marketers assume, and how often depends heavily on which engine you ask. In a 2026 analysis of 34,234 AI answers by AI-visibility platform Leapd.ai, ChatGPT named a brand in about 0.59% of responses, Perplexity in about 13.05%, and Grok in about 27% — roughly a 46-times gap between the least and most brand-naming engines. This reference lays out the figures, attributed and dated, explains why the engines diverge so sharply, and separates "naming a brand" from the related-but-different questions of being cited as a source and being recommended.
Last reviewed: 24 August 2026. The brand-naming rates below are from Leapd.ai's 2026 analysis of 34,234 AI responses. It is a single-vendor, directional study whose full per-engine methodology is not published, and brand-naming rates are highly sensitive to the mix of commercial versus informational prompts in the sample — so read the ordering and order-of-magnitude as the finding, not the second decimal place, and cite "Leapd.ai, 2026" with the date when you reuse the numbers.
How often does each AI engine name a brand?
Rarely on ChatGPT, occasionally on Perplexity, and most often on Grok, on the reported sample. The headline is the spread, not any single figure:
| Engine | Share of answers that named a brand | Notes |
|---|---|---|
| ChatGPT | ~0.59% | Answers many prompts from training data; retrieves selectively |
| Perplexity | ~13.05% | Performs a live web search on essentially every query |
| Grok | ~27% | Highest brand-naming rate in the sample |
Source: Leapd.ai, 2026 (analysis of 34,234 AI responses). The gap between the lowest and highest is about 46 times — a far wider divergence than the same brands would see across classic search engines. The practical takeaway: there is no single "AI brand-mention rate." A brand that appears often in Perplexity answers can be nearly absent from ChatGPT's prose for the same questions, and a one-engine spot check will badly misread total visibility.
Why do the engines differ so much?
Because they retrieve differently. The brand-naming rate is downstream of when and how each engine goes to the live web:
- ChatGPT answers many prompts from its trained parameters and performs a live search only for a subset. Leapd.ai reports it retrieves for roughly 53.5% of commercial queries versus about 18.7% of informational ones. When it does not retrieve, it tends to describe a category rather than name a vendor — so a brand name surfaces only when retrieval fires and a specific brand is the answer.
- Perplexity performs a live web search on essentially every query and foregrounds named sources, which mechanically surfaces more brand names in the visible answer.
- Grok named brands most often in the sample; treat its high rate cautiously given the single-vendor source and the smaller likely sub-sample per engine.
The pattern is that an engine that searches the live web more often, and shows its sources more prominently, names more brands. This is the same architecture split the corpus has documented from the other direction: engines cite largely different sets of pages, because they retrieve from different places at different times.
There is no single number for how often AI names your brand. The rate swings roughly 46-fold across engines, so "our AI mention rate" is only meaningful once you say which engine, for which kind of question.
Is naming a brand the same as citing a source?
No — and conflating the two is the most common measurement error here. Three distinct things happen in an AI answer, and each is measured separately:
| Metric | What it measures | Example |
|---|---|---|
| Brand-naming rate | Whether a brand name appears in the visible answer text | "Tools like Acme and Globex do this" |
| Source citation | Whether a domain is cited as the source of a claim | A footnote linking acme.com |
| Recommendation | Whether the answer actually endorses the brand | "The best option is Acme" |
These come apart constantly. An engine can cite your page as a source and still recommend a competitor; it can name a brand it never cited; and it can lift a fact from your domain without ever printing your name — a ghost citation. So a low brand-naming rate is not the same as being invisible. It is one lens among several, and it pairs naturally with the brand-mention gap versus source gap audit, which separates "the engine never names us" from "the engine never cites our pages" because they need opposite fixes.
This is also why brand-naming rate does not contradict the source-count data. ChatGPT cites on the order of 15 source domains per answer on Semrush's 2026 figures, yet names a brand in under 1% of answers here — because citing many domains as background sources is a different act from printing a vendor's name in the sentence a reader reads.
Does this square with ChatGPT's shift toward first-party sources?
Yes, cleanly. We reported separately that ChatGPT is resolving more of its citations to first-party domains after GPT-5.6 — using the site: operator to fetch named domains directly. That is about which domains it cites as sources, and it can rise at the same time as the brand-naming rate in prose stays low. The model can quietly fetch and quote a brand's own page while still writing an answer that describes the category without naming vendors. Both are true: ChatGPT increasingly grounds on first-party pages as sources, and it rarely names brands in the visible text. They are two different surfaces of the same answer, and a brand should measure both rather than assume one implies the other.
What should a brand do about it?
Stop chasing a single "AI mention rate," and work the engine that matches where your buyers actually ask:
- Measure per engine, per question type. A 46-fold spread means one blended number hides everything. Track ChatGPT, Perplexity, Gemini, Grok, and Google separately, split by commercial versus informational prompts, the way an AI visibility audit should.
- Win the naming engines with corroboration, not self-declaration. On Perplexity and Grok, which surface live sources, brand names ride in on third-party pages and clean first-party facts — the same levers behind why AI cites one brand and how to get recommended in ChatGPT.
- On ChatGPT, be the first-party source it fetches. Because it names few brands but reaches for named domains, make your specs, pricing, and official facts clean, server-rendered, extractable text so that when it does retrieve, your page is what it lifts.
- Track the trend, not the snapshot. Brand-naming rates move with each model release. A quarter-over-quarter line per engine tells you far more than any one reading.
The honest read for late 2026: how often AI names your brand is real, worth tracking, and almost meaningless as a single blended figure — it only becomes decision-grade once you split it by engine and question type and watch it over time. That per-engine, over-time view is exactly what Buffy Intel is built to measure. Questions: [email protected].