The 181-Brand AI Visibility Study

What the 181-Brand AI Visibility Study found, engine by engine

No engine was meaningfully kinder than another. Perplexity left 171 of 181 brands absent and Gemini, the most generous, still left 164. That is a spread of three percentage points across four engines built by different companies on different indexes, which means a brand missing from one is almost always missing from all of them rather than losing a single platform.

1. Perplexity 94% absent
2. ChatGPT 93% absent
3. Google AI Overviews 93% absent
4. Gemini 91% absent

• Absent
• Mentioned
• Cited

How invisible, exactly?

The first pass also scored each brand from 0 to 100 on how strongly it surfaced. The mean was 18.4 out of 100, and 61 brands (34%) scored a flat zero. Only three brands in the entire sample scored above 50.

Methodology

We built a sample of 181 United States ecommerce brands that were actively paying for Meta ads, wrote one buying question per brand in that brand's own product category, and put the same question to four AI engines. An answer counted as a hit only when the engine named the brand or cited its domain. A brand counts as absent only when all four engines returned nothing.

Sample

181 US ecommerce and DTC brands

Selection

Active Meta advertisers running their own storefront

Engines

ChatGPT, Perplexity, Gemini, Google AI Overviews

Coverage

All four engines queried for all 181 brands

Query

One unbranded "best [category]" buying question per brand

Classification

Cited, mentioned or absent, per engine

Sample built

2026-06-15 to 2026-06-17

Engines queried

2026-06-18 to 2026-06-25

Published

8 August 2026

Licence

CC BY 4.0, reuse with attribution

How the sample was built

We pulled advertisers from the Meta Ad Library between 2026-06-15 to 2026-06-17, looping a list of consumer product keywords and keeping brands that ran their own online store. We removed tracking domains, duplicates and large-retailer subdomains. That left 181 clean United States ecommerce and DTC brands. Advertising was the qualifying signal, because a brand paying for traffic has already decided that being findable is worth money.

What the sample looked like

Category is the brand's own Facebook page label across 69 distinct values, and it is weak data: 46 of the 181 brands used a label that describes nothing, such as "Brand", "Website" or "Product/service". We report it for shape only and drew no conclusions from it.

What we asked the engines

For each brand we wrote one buying question in that brand's own category, phrased the way a shopper would type it, in the shape of "best [category]". We did not use the brand's name. The point was to see whether an engine reaches for the brand unprompted when a customer is choosing what to buy, which is the moment the sale is decided.

How each answer was classified

• Cited: the brand's own domain appeared as a source behind the answer.
• Mentioned: the brand was named in the answer text but its site was not cited.
• Absent: neither. The engine answered the question and the brand did not exist in it.

Four engines were queried for every brand between 2026-06-18 to 2026-06-25: ChatGPT, Perplexity, Google AI Overviews and Gemini. A brand counts as absent overall only when all four returned absent.

Why some of our older material says 72%

The first pass, run 2026-06-15 to 2026-06-17, used a single engine and found 131 of 181 brands absent, which is where the 72% figure came from. Going back over the same 181 brands with all four engines moved the number up to 158 (87%). Two brands that looked absent turned out to be visible; twenty-two that looked visible turned out to be absent once we checked properly. We now lead with the four-engine number and keep 72% on record as the conservative first-pass floor, which is the figure our earlier videos quote.

Limitations

This study measures one specific thing: whether four AI engines named or cited 181 advertising US ecommerce brands on one unbranded category question, during one week in June 2026. It is not a random sample of ecommerce, not a longitudinal study, and not a measure of what those engines would say today. The limits below are stated so nothing here reads as more certain than it is.

• This is not a random sample of ecommerce. Every brand was advertising on Meta, in the United States, in June 2026. That selects for businesses with a budget. Do not read 87% as a fact about all online stores. It is a fact about these 181.
• One question per brand per engine. Visibility is query-specific. Several brands that were absent for the broad category question turned out to be cited for a narrower one when we followed up. A different question would produce a different number.
• One run per question. AI answers are not deterministic, and we did not sample the same question repeatedly to measure variance, so treat individual brand verdicts as a snapshot rather than a settled fact.
• June 2026, and engines change fast. Any of these results could move within weeks of publication.
• Gemini is the weakest leg. It was queried on a free tier subject to rate limiting, and a rate-limited response can come back empty. An empty response misread as silence would inflate the absent count, so the collector was changed to stop rather than record a false absent. It is still the engine we would trust least here.
• Category labels are self-reported and messy, as described above. Eight of the 181 were also later found not to be genuine online sellers.
• We are not neutral. We sell the fix for this problem. That is exactly why the method is written out in full: so you can disagree with it, or run it yourself.

What we take from this AI visibility study

Every brand in this sample was paying for Meta advertising when we built it, and 158 of 181 were still absent from all four engines. Paid reach on social did not carry these brands into AI answers about their own category. Only 15 had their own site cited by even one engine. The study recorded whether each brand was named or cited and did not measure why, so we draw no conclusion here about what those 15 have in common.

If you want to check your own brand against this, the Free AI Visibility Audit runs a similar check on your category across ChatGPT, Perplexity, Gemini and Copilot, or you can follow the steps in how to run an AI visibility audit yourself and do it by hand. The mechanics of getting cited are covered in how to get cited by ChatGPT.

This study covers ecommerce brands. Our real estate research puts the same kind of question to the engines about agents and brokerages: start with the 500-query Houston study and the 150-query real estate study, or use the real estate SEO statistics list, which sets those findings beside third-party figures.

Press kit: cite, download and reuse this study

Everything here is free to republish under CC BY 4.0, including commercially, with attribution and a link back. The underlying row-level data is published alongside the charts so anyone can check the figures rather than take them on trust. No permission request is needed and there is no embargo.

Download

• CSV

  Full dataset, CSV

  All 181 rows, anonymised: category, platform, each engine's verdict, visibility score. 13 KB

• PNG

  Headline chart, PNG

  The 87% finding. 1200px wide, source line and licence in the image

• PNG

  By-engine chart, PNG

  Absence across all four engines. 1200px wide, source line and licence in the image

Brand names and domains are withheld from the dataset. The brands did not consent to being named and naming them would publish a commercial weakness they were never asked about. Every aggregate on this page recomputes exactly from the rows provided.

How to cite it

For journalists

Questions about this study

The questions below are the ones we are asked most often about the sample, the engines and the 87% figure itself, including why some of our earlier material quotes 72% instead. If something is not answered here, the raw dataset is downloadable above and we answer methodology questions directly.