Blog · 23 September 2026 · By HeardOf

Named but not cited: being mentioned by an AI is a different job from being its source

Is being named in an AI answer the same as being cited by it?

No. They are two counts, and in our run of 4 September 2026 they came apart: across 240 brand-answer pairs — six brands in each of four B2B software categories, ten answers each — a brand was named in the answer text in 45.0% of pairs, and its own site was among the answer's sources in 30.0%.

Named means the brand's name appeared in the text of the answer, links the engine printed inline included. Cited means at least one URL the engine returned as a source for that answer was on the brand's own domain. An answer that names Samsara and lists twenty sources, none of them on samsara.com, has named Samsara and not cited it — Perplexity returned exactly that to the small-business fleet prompt on 4 September 2026. Both counts come from the same stored answers — ChatGPT and Perplexity, five buyer prompts per category, run on 4 September 2026 and re-counted on 23 September 2026. How five vendors define AI visibility, and what each one divides it by, is in what AI visibility is, and how it is measured, which quotes their own pages as read on 24 September 2026.

The two counts do not move together, and the gap has a direction:

Two rows are the point. The pairs that came apart on 4 September 2026 almost all came apart one way: 37 named without a citation, 1 cited without a name. And the engines differ on how far: ChatGPT named a brand without citing it in 11 of its 60 mentions and never cited a measured brand's site without naming it; Perplexity named without citing in 26 of 48 — more than half the times it wrote a measured brand's name that day, none of its sources was the brand's site.

How often does an AI answer name a brand and link to someone else?

In 21 of the 40 answers, at least one measured brand was named with no source on its own site (37 of the 108 named pairs), 4 September 2026. The split that day was 6 of ChatGPT's 20 answers and 15 of Perplexity's 20. In 25 of those 37 pairs the same answer did cite a measured competitor's own site, in the same run: the brand was in the text and a rival was in the sources.

What the sources were instead, across those 21 answers of 4 September 2026: reddit.com in 11 of them, all Perplexity; forbes.com in 7; a review site — G2, Capterra, GetApp, Software Advice or SourceForge — in 7. On Perplexity that day, 18 of the 26 named-only pairs sat in an answer that also cited a Reddit thread.

Two answers show the shape. ChatGPT, asked on 4 September 2026 for the best GPS fleet tracking tool for small fleets in 2026, named Motive, Samsara and Azuga and returned eleven sources, none on any of the six measured fleet vendors' domains; five of the eleven were techradar.com pages. Perplexity, asked the same day "Follow Up Boss vs kvCORE vs Propertybase — which is better?", named all three, returned sixteen sources, and not one was on followupboss.com, insiderealestate.com or propertybase.com; three were on ustechautomations.com and two on selecthub.com. A head-to-head question about three vendors, answered on 4 September 2026 without a page from any of them.

Which brands were named but not cited?

Samsara, kvCORE and Follow Up Boss most often, in 4 of their 10 answers each, then Teamwork.com, Azuga, Verizon Connect and Propertybase at 3 each, all on 4 September 2026. Two of the four categories no earlier post of ours has shown brand by brand — creative agency project management and fleet management — carry the pattern in full:

Read Teamwork.com against Productive, 4 September 2026: named 10 times against 8, cited 7 against 8, and Productive's 8 citations fall in exactly the 8 answers that name it. Four brands named at least once that day — Productive, Fleetio, Scoro and Top Producer — were named and cited in the same answers every time, both or neither. Of the 24, fifteen were named more often than cited, eight were equal — four of those at zero on both — and one, Workamajig, was cited more often than named, all re-counted on 23 September 2026.

Can a brand be cited without being named?

Once in 240 pairs, on 4 September 2026. Perplexity, asked for the best PM software for agencies with billable hours and retainers, listed three workamajig.com blog posts among its twenty sources and did not write the word Workamajig anywhere in the answer, 4 September 2026. The same answer, that day, named and cited Teamwork.com and Productive. In the other nine agency answers of 4 September 2026 Workamajig was neither named nor cited.

Add the fifth category, dental, from the same 4 September 2026 run, and the number of brands cited more often than named goes from one to three. The dental raw answers are no longer held — who the AI actually cites explains why — but the per-brand summary written on the run day, 4 September 2026, survives: Curve Dental named in 7 of 10 answers and cited in 9, Denticon named in 4 and cited in 5. Across all 30 brands that day: 3 cited more often than named, 18 named more often than cited, 9 equal. The dental pair was already the worked example in how to get your SaaS cited by ChatGPT and Perplexity; what the four re-countable categories add is how rare it is — one pair in 240 — and that in each of the four, named-without-cited outnumbered cited-without-named: agency 5 to 1, fleet 12 to 0, real estate 15 to 0, staffing 5 to 0, all on 4 September 2026.

I show up in ChatGPT, so why hasn't anything changed?

Because "showing up" is most likely the count you can see by reading the answer — your name — and that is the larger of the two counts, by 36 pairs in the run of 4 September 2026. On ChatGPT that day the two were close: 49 of the 60 mentions came with a link to the brand's own site. On Perplexity that day they were not: 22 of 48.

What a mention or a citation does for you afterwards we cannot measure, and the earlier post says the same about clicks; a citation count is not a traffic report. What the stored answers do show is the shape of the two things. A mention is words in the text. A citation is a URL the engine returned beside the text — one of 5 to 12 per answer on ChatGPT in the run of 4 September 2026, one of 16 to 20 on Perplexity. The first tells you your name was in the answer; the second, that a page of yours was among its sources. On 4 September 2026 the first was true without the second 37 times, and the second without the first once. On the Brand24 and Mention pages we read on 25 September 2026, the mention feed collects your name from posts, articles and reviews, and Brand24 tracks AI answers through a second product, Chatbeat; LLM brand monitoring vs social listening sets out what each one watches.

They are two jobs because they moved separately. On 4 September 2026 Samsara and Fleetio were named in 9 and 8 answers and cited in 5 and 8; Fleetio's site was a source in every answer that named it, Samsara's in 5 of 9, and 3 of the 4 it lost were Perplexity's. If what you checked was your name and what changed was nothing, the count you did not check is the one that went to a rival's page in 25 of the 37 pairs where the two came apart that day. Which of your pages an engine reaches for, engine by engine, is the earlier post's subject, and this one does not repeat it.

How was this measured?

On 4 September 2026 we asked ChatGPT and Perplexity five buyer prompts in each of four B2B software categories — creative agency project management, fleet management and GPS tracking, real estate CRM, staffing ATS and recruitment CRM — with six measured brands per category, and stored every answer with every URL the engine returned as a source. Each answer from that day was then scored once per brand: 240 pairs. Every figure above was re-counted from those stored answers on 23 September 2026.

The two rules, in plain words, as applied on 23 September 2026. A brand is named if its name, or one of the alternative names we listed for it, appears in the answer text as a whole word, in any case, in our re-count of 23 September 2026 — Motive also as KeepTruckin, kvCORE also as Inside Real Estate or BoldTrail, Teletrac Navman also as Teletrac or Navman. In that re-count, the answer text is everything the engine wrote, including any link text or web address it printed inline. In the same re-count, a brand is cited if at least one URL among the answer's sources has the brand's registered domain as its host, with www. removed, or a subdomain of it — help.fleetio.com counts for Fleetio, and kvCORE's registered domain is insiderealestate.com. Each rule, as we applied it on 23 September 2026, is yes or no per brand per answer; a brand named three times in one answer counts once, and so does a brand cited three times.

Two lines the rules draw are worth showing, both from the run of 4 September 2026. ChatGPT once cited a page at propertybase.lwolf.com that day, in an answer that named Propertybase; that host is not propertybase.com, so the pair counts as named and not cited. Re-counted on 23 September 2026 under a rule that read it as Propertybase's site, Propertybase would be cited in 1 of 10 answers rather than 0, the named-only count would be 36 rather than 37, and the cited rate 30.4% rather than 30.0%; no brand's place in the tally would move. The other line runs the opposite way. In one ChatGPT answer about staffing software on 4 September 2026, the only occurrence of JobAdder is inside a web address the engine printed in the text — a LinkedIn PDF whose filename contains the name — and the name rule counts it as named. Read as a reader would, that pair is not a mention, and re-counted that way on 23 September 2026: the named count would be 107 rather than 108, or 44.6% rather than 45.0%; the named-only count 36 rather than 37; ChatGPT's mentions 59 rather than 60, with 10 named-only rather than 11; and JobAdder would read 4 named and 4 cited, in the same four answers, moving the 24-brand tally to fourteen named more often than cited, nine equal and one cited more often than named and the 30-brand tally to seventeen named more often than cited, ten equal and three cited more often than named.

The fifth category, dental, is here only as six rows of the 30-brand tally. Its stored answers no longer exist, for the reason the earlier post gives, so its figures come from the per-brand summary produced on 4 September 2026 and cannot be re-counted. The four categories' 24 rows in that same summary match our re-count of 23 September 2026 brand for brand, which is why we trust the dental rows.

Everything this post says about our own run — the two counts, the rules, the examples, what was lost — comes from our own stored answers as re-read on 23 September 2026, and carries that date or the run date in the sentence that makes it. What the closing below says about the audit we sell comes from our own description of it in which AI engines we check, as that post read on 23 September 2026, and from nothing newer. That is the half of this post a reader cannot check, and we are saying so rather than letting you find the seam. Everything else is a description of the addresses the engines returned that day, which we hold and have not opened, or a link to an earlier post of ours. Where we did not check, the text says so.

LLM brand monitoring vs social listening: what each one watches