The AI Shortlist: Marketing Tools

When a buyer asks an AI engine which marketing tool to use, the engine returns a shortlist of three to five names. That shortlist is now the top of the funnel for an entire category, and almost nobody measures their position in it. So we measured it.

In July 2026 we probed four AI engines — ChatGPT, Perplexity, Gemini, and Claude — with the questions marketing-tools buyers actually ask. Sixteen answer-probes per brand, across six brands in the category. Separately, we captured how many live ads each of those brands was running. Setting the two datasets side by side produces the finding that matters: they have almost nothing to do with each other.

The Shortlist

How often each brand was named, out of 16 answer-probes: Semrush 11. Jasper 11. Wynter 10. Mailchimp 8. Copy.ai 2. Lytms 0 (of 13 probes).

We include our own zero deliberately. We are a young brand with almost no third-party corroboration, and the engines reflect that accurately. A measurement that flatters the people publishing it is not a measurement.

Across those answers, the names the engines reach for most often in this category are not always the brands you would expect from category marketing: Ahrefs (15 mentions), HubSpot (15), Qualtrics (14), ActiveCampaign (13), Constant Contact (12), Klaviyo (10), Brevo (9), SurveyMonkey (8). The engines answer with the brands the web has written most about — which is not the same population as the brands spending the most on demand generation.

The Finding: Ad Volume Does Not Buy AI Presence

Set the live-ad counts against the citation rates and the independence is stark. Copy.ai was running 40 live ads at capture and is named in 2 of 16 answers. Wynter was running 2 live ads and is named in 10 of 16 — five times the answer presence on a twentieth of the ad volume. Semrush, at 59 live ads, lands at 11 of 16; Jasper, at 40, also lands at 11.

The pair to sit with is Jasper and Copy.ai. They are direct competitors, comparable in category position, both running dozens of live ads at capture — and one appears in eleven answers while the other appears in two. Whatever separates them, it is not paid demand generation, because on that axis they are matched.

This is not an argument against advertising. It is an argument that the two systems have different currencies. Paid placement is bought at auction, in cash, and stops the moment the budget stops. Answer presence is earned through the corroborating record an engine reads — reviews on the aggregators, inclusion in the comparison articles that rank, community threads, documentation the web quotes back. That record accrues slowly, cannot be purchased directly, and does not disappear when a campaign pauses.

The practical consequence for a marketing team: your ad account tells you nothing about your position in the answer layer, and no amount of spend will move it. They require separate work, measured separately.

What the Zero Actually Means

A brand named in 0 of 16 answers is not being judged by the engines and found wanting. It is absent from the sources the engines read when they assemble an answer. That is a different problem with a different fix.

Engines synthesize from a retrieved corpus. A claim that exists only on your own domain is a self-claim and carries little weight in that synthesis; the same claim repeated across a review aggregator, a ranked comparison article, and a community thread becomes a fact the engine will assert. Absence from answers is almost always absence from that corroborating record, not a verdict on the product.

It also compounds. Answers feed the next crawl, and early consensus is sticky: once a corpus repeatedly states that the leaders in a category are A, B, and C, displacing that list takes considerably longer than earning a place in it did while the category was still forming.

Method and Limitations

Engines probed: ChatGPT, Perplexity, Gemini, Claude. Four category buyer questions per brand, asked identically to each engine — 16 answer-probes per brand. "Cited" means the brand name appears verbatim in the answer text; no fuzzy matching, no inference. Citation probes were run July 29-30, 2026. Live-ad counts are separate captures taken July 4-19, 2026, combining a brand's public ad-library total with its public Google ad-transparency listings.

Limitations, stated plainly. AI answers are non-deterministic: the same question asked twice can return different names. In this very dataset we scanned one brand twice roughly two hours apart and its citation count moved from 9 to 12 out of 16 — which is precisely why these figures are point-in-time rates rather than fixed scores, and why the series re-runs quarterly rather than claiming a permanent ranking. Six brands is a small cohort; every claim above is scoped to the brands we probed, not to the category as a whole. The ad counts and the citation probes were captured on different dates, so the comparison describes two measured states of the same brands weeks apart, not a controlled experiment. Every number here is a real retrieved value; nothing is modeled, extrapolated, or estimated.