Perplexity's share of AI chatbot referrals nearly halved in two months, and the one big assistant we do not audit now sends slightly more traffic than one we do.

StatCounter's August figures, read on its own page: ChatGPT 79.4%, Gemini 10.9%, Perplexity 4.31%, Copilot 2.79%, Claude 2.57%. Jarboe, citing the same source, gives June as Perplexity 7.91% and Gemini 7.94%, so Perplexity has nearly halved in two months while Gemini rose by about three points. He also cites Similarweb's May chatbot visit share with Perplexity and Copilot at 1.3% each. His conclusion is "De-weight it" rather than "delete it", and he warns that a pooled score across engines "can create a false sense of precision."

Limits: StatCounter measures referrals from sites carrying its tracker, worldwide and across every kind of site, not coaching buyers. The June figures come from Jarboe, not a first-hand read.

Verbatim, translated: "We have not observed any sudden decline in our audiences, aside from seasonal effects." This comes about seven weeks after AI Overviews reached France. He says visitors arriving from ChatGPT convert at rates "26 times higher" than visitors from Facebook and "260 times higher" than those from Discover, and that July and August each brought 24,000 new subscribers.

Limits: one subscription newspaper, relative rates with no base figures, no breakdown of Google search referrals, and nothing on how often Le Monde appears inside an AI Overview.

Older, and new to this desk. A review of 45 studies. Its central line, verbatim: "no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability or downstream behavior." Commercial audits in the review showed "low source overlap, substantial run-to-run variability, and persistent fidelity gaps". Only topical relevance and position in the context showed reproducible effects, and citation-focused rewrites could make retrieval worse. The author asks for "repeated measurements, paraphrases, controls, human validation, and multi-actor interference."

Limits: one author, a survey of other people's studies rather than new measurement, read at the abstract level today.

A survey of 1,042 content marketers. Verbatim: "14% report 'strong results'", the lowest in the survey's twelve years. "92.4% now use AI for blogging." And: "using AI does not improve content performance." Orbit also notes that "traffic goes down and conversion rates go up."

Limits: self-reported by marketers, and "strong results" is their own judgment.