The Brand Invisibility Report: 7,048 AI answers, measured

Maark original research · May 22 – July 29, 2026

Ask an AI engine for a product recommendation — "best walking pad for a standing desk," "magnesium supplement for sleep" — and it answers with confidence. Somebody's products fill that answer. We wanted to know, with real measurement rather than anecdote: is it yours?

We operate AI-visibility measurement infrastructure for the brands on our platform. Between May 22 and July 29, 2026, that infrastructure stored 7,048 probes of five AI answer engines — ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode — across 1,085 distinct buying-intent prompts tracked for 21 brands. This report is what the corpus says.

The headline: the shelf is empty

Of our probes, 6,914 were non-branded — buying-intent phrasing ("best X," "X for Y") that never names a brand. Across those 6,914 answers:

• The measured brand was mentioned in 153 answers (2.2%).
• The measured brand's pages were cited as sources in 266 answers (3.8%).
• In 6,752 answers (97.7%), neither the measured brand nor any of its tracked competitors appeared.

One precision note we hold ourselves to: 97.7% does not mean "no company was named." AI shopping answers are full of names — retailers, publishers, products outside each brand's tracked set. What we measured is the competitive set each brand actually cares about: itself plus the rivals it tracks. In 97.7% of non-branded buying-intent answers, that entire set was absent. The conversation happened without any of them.

Your competitor isn't beating you. Nobody is.

The instinctive fear is "if I'm not in the answer, my competitor is." The data says otherwise. Outcomes where the measured brand was absent but a tracked competitor was named: 9 of 6,914 probes — 0.13%.

When you're missing from an AI answer, your rival almost certainly is too. This isn't a lost shelf; it's an empty one, for the whole tracked set at once. Whoever fills it first isn't displacing an incumbent — for most of these prompts, there isn't one yet.

Ask by name, ask by need: 80% vs 2%

The corpus includes 134 branded probes — prompts that name the brand directly ("is [brand] good for X"). The brand was mentioned in 107 of them (79.9%).

That's the structural finding under everything else. The engines know these brands. Ask by name and they surface the brand roughly 80% of the time. Ask by need — the way a buyer who hasn't heard of you asks — and it drops to roughly 2%. The gap isn't awareness inside the model; it's retrieval at the moment of commercial intent. The engines can find you. They just don't reach for you when it counts.

You can be the evidence without being the recommendation

A quieter oddity: brand pages were cited more often than brands were mentioned — 3.8% cited vs 2.2% mentioned. Engines will quote your page as a source for an answer that never names you. Your product specs and your how-to content feed the answer; the recommendation goes unattributed, or goes to the content that packaged the comparison.

Per engine, the cited-over-mentioned pattern holds on four of the five; ChatGPT is the exception, mentioning slightly more often than it cites.

The engines mostly agree

| Engine | Non-branded probes | Brand mentioned | Brand pages cited | Branded-prompt mention rate | |---|---:|---:|---:|---:| | Perplexity | 1,626 | 2.0% | 6.3% | 80.6% | | ChatGPT | 1,618 | 1.9% | 1.1% | 63.9% | | Gemini | 1,550 | 2.5% | 5.4% | 85.7% | | Google AI Overviews | 1,447 | 2.1% | 2.3% | 88.9% | | Google AI Mode | 673 | 3.3% | 4.3% | 100.0%* |

*Google AI Mode's branded-probe sample is small; read that 100.0% as directional, not precise.

No engine breaks the pattern. Non-branded mention rates sit in a narrow 1.9–3.3% band across all five. This is not a ChatGPT quirk or a Google quirk; it's how the current generation of answer engines handles commercial queries. Perplexity and Gemini cite brand pages at the highest rates (6.3% and 5.4%), which makes them the most reachable surfaces for brands with citable content.

What fills the answers instead

If brands aren't in the answers, what is? We normalized every citation in the corpus into a graph: 70,106 citations across 15,982 distinct hosts. Our type classification currently covers about 47% of that graph; of classified sources:

• Blog / editorial: 45.9%
• Video: 15.6%
• Reference: 14.3%
• UGC / community: 8.2%
• News: 5.9%
• Social: 4.2%
• Review marketplaces: 3.1%
• Docs: 2.1%
• Directories: 0.8%

The reading is blunt: engines cite content about products — guides, comparisons, videos, community threads — not product pages. Blog and editorial content alone accounts for nearly half of classified citations; add video, reference material, and community discussion and you're past 80%. The path into AI shopping answers runs through editorial formats, not through your catalog.

Method

• Corpus: 7,048 stored probes of five AI answer engines (ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode), collected May 22 – July 29, 2026, refreshed by a daily automated sweep.
• Brands: 21 brands measured on the Maark platform (Maark/Nordica client projects), skewing ecommerce and health/wellness.
• Prompts: 1,085 distinct tracked prompts in buying-intent phrasing ("best X," "X for Y"), split into non-branded (6,914 probes) and branded (134 probes) sets.
• Per-probe record: whether the tracked brand was mentioned in the answer text, whether any of its pages was cited as a source, and which tracked competitors were mentioned.
• Citation graph: all answer citations normalized to hosts — 70,106 citations across 15,982 distinct hosts — with source-type classification covering roughly 47% of the graph to date.

All percentages in this report are computed directly from this corpus as of July 29, 2026.

Limitations

We hold research to the standard we'd want from anyone else's, so the caveats are part of the finding:

1. Single-vendor capture pipeline. Probes are API-based measurement, with no logged-in personalization. Real user sessions may differ.
2. Cohort skew. 21 brands, weighted toward ecommerce and health/wellness. This is a fleet measurement, not a population study; cohort sizes are disclosed throughout.
3. Prompt selection. Prompts are our tracked buying-intent prompts, not a random sample of real user queries.
4. "Competitor" is bounded. It means each brand's tracked competitor set, not every brand on earth — hence the precision rule on the 97.7% figure.
5. Point-in-time percentages. The structural findings — the branded/non-branded gap, cited-over-mentioned, editorial dominance — are robust across the corpus and the full window (cited-over-mentioned holds on four of five engines). The exact percentages will move, because AI answers churn: external research (Authoritas) puts AI-answer citation churn at roughly 70% over 2–3 months.

What we're measuring next

Three extensions are already running: pushing citation-type classification past the current roughly 47% of the graph; widening the brand cohort beyond the ecommerce/wellness skew; and longitudinal churn measurement on our own corpus — how long a brand that enters an AI answer actually stays there, and which content formats got it in.

The 97.7% number will not hold forever. These answers are being filled right now, mostly by editorial content, and the churn data says the shelf gets restocked every few months. That is either a threat or the most open distribution channel in commerce, depending on whether you're measuring it.

Maark measures it continuously — brand mentions, page citations, and competitor presence across all five engines, per brand, refreshed daily. If you want to know your own numbers before your category fills in, that's what we built it for. Join the waitlist.