// SIGNAL
AI & SEO Research
Signal is our research program. Every study is public, every number has a methodology, and we run every playbook on ourselves.
Publications
Original studies we ran.
Who Decides If Your Exchange Is Safe? ChatGPT Asks You. Everyone Else Asks Reddit.
On the highest stakes question a crypto buyer types, ChatGPT builds the answer out of the exchange's own website about half the time. Every other engine reaches for Reddit. 1,300 engine calls, then replicated a week later.
Readiness Alone Doesn't Earn Citations. Authority Does.
We measured 45 crypto and fintech brands across every engine and controlled for brand size. The result changed what we sell.
95.5% of Crypto and Fintech Sites Are Open to AI Crawlers. Access Isn't the Gap.
We audited 146 crypto and fintech sites across nine dimensions. Whatever keeps them out of answers, it is not a locked front door.
Analysis
Method, and a hard read of the numbers everyone else is quoting.
Every Guide to Picking an SEO Agency Says What to Ask. None Says How to Score the Answer.
Seven questions, scored out of 14, so two people can interview the same agency and land on the same number. We checked 17 agencies selling AI visibility work: two publish a price, and not one publishes how many times it runs a prompt.
Reddit's ChatGPT Citations Fell 86%. Last Year They Fell 95% And Came Back.
A six point post about ChatGPT rebuilding its sources is circulating. One claim describes a measurable event. We traced the other five back to unrelated studies with swapped denominators, moved dates and, in one case, no source at all.
One Question Told Us Something Dramatic. Thirty One Told Us It Was Wrong.
We pointed the method at our own instrument. The pilot cleared a threshold we had fixed before collecting, the full study did not, and the two metrics disagreed. So we published the disagreement and changed nothing.
Gemini Cites You. Your Report Credits Medium.
Gemini cites through a redirect that names the platform, not the publisher. If your content lives on somebody else's domain, your visibility report is crediting them. 595 citations measured across two topic sets.
Most AI Visibility Numbers Move Because Someone Changed the Questions. The Fixed Set Method Doesn't.
The Fixed Set Method is how we measure generative engine optimization without lying to ourselves. A fixed prompt set, multi run sampling, and a nine check readiness score.
llms.txt Is on One in Ten Sites and Pulls 0.1% of AI Crawler Traffic.
Everyone is generating the file. No major vendor has committed to reading it, and the crawler logs show almost none do.
AI Reshaped Search in 2026 Faster Than Anyone Can Measure It
Half of 2026 is gone and AI has moved search traffic, brand discovery, and even software valuations. The attribution gap is the real story.
Source audits
Where the numbers everyone quotes actually came from.
We Opened the Studies Behind Seven Numbers Everyone Quotes. Three Have No Public Dataset.
Seven of the most repeated figures in AI search optimization, traced back to source. Three come from datasets published nowhere, two are quoted at the wrong scope, and two hold up exactly as stated.
Four Numbers We Published and Had to Correct.
A correlation read off a compressed range, a rubric that scored a two word page 54 out of 100, a method that flattered sites we could not fully measure, and a study we retracted three times in one day. Ours, not somebody else's.
The 76% Top 10 Figure Is Real. The Same Publisher Now Reports 38%.
Ahrefs measured 76% in July 2025 and 38% since. Its cleanest measurement of the question people actually mean puts it at 12%. Which number answers which question.
Reddit Is About 12% of What AI Answers Cite, Not 54%.
The 54% to 71% figure is Reddit's share of user generated content, not of all sources. Multiply it by the UGC share and Reddit lands near 12%, in three independent datasets.
82% of AI Citations Are Third Party, but That Is a Share of Source Slots.
Aleyda Solis measured the top 10 cited source domains per brand across 45 brands. Read as a share of citations it overstates the case. Owned domains hold about 10% to 11%.
85% of AI Brand Mentions Come From Third Parties, at the Top of the Funnel.
AirOps counted 21,311 brand mentions in October 2025. The figure is real, dated and open. It measured discovery queries rather than purchase intent ones, and that funnel stage is the whole story.
84% Earned Media Is a Bucket Defined by Exclusion.
Muck Rack measured 84% across more than 25 million links in May 2026. Journalism, the thing a PR retainer actually buys, is a small slice of that bucket.
The Trustpilot 1% to 75% Gap Shrinks Once You Match the Brands.
The headline comparison came from unmatched tiers in a study Trustpilot commissioned. The authors then built a matched cohort of 206 brands per tier to control for brand strength, and that is the number worth reading.
Three Findings Everyone Repeats Exist Only Inside the Articles Reporting Them.
The Consensus Gap, the 44.2% ski ramp and the 74% shortlist figure. No public dataset, no method note, no way to check any of the three.
Get the next study.
One email per study. That’s it.
Want this run on your brand? The Diagnostic applies the same methodology to your buyers’ questions. $7,500, six weeks. Roadmap at week three, corrections shipped and re-measured by week six.
Get the Diagnostic →