Google gets better at blocking the tools that read its results, and Perplexity's cutover is eight days out.

Several third-party tools that read Google's results report a sharp drop in what they can collect since around 13 September. Nozzle's Derek Perkins reported "about an 80% drop in data he was able to get from Google" and said DataForSEO showed the same problem. Sistrix's status page says "our data collection is currently running at a reduced rate", and Sistrix's Steve Paine said Google "has been continuously changing how search results are delivered for about a year now, currently at a very high pace." Ahrefs and Semrush were affected to varying degrees. There is no Google statement, and Schwartz warns that tool data from this period may be wrong. SerpApi's own public metrics page, read 19 September, shows its Google AI Overview API at a 99.875% success rate over the last 30 days, which measures whether a request returned, not whether an overview came back.

Method: 129 everyday-question transcripts from a GPT-5.4 search agent using Exa as the search provider, 113 matched pairs of pages supporting the same fact (103 confirmed by blinded human review), replayed with the pair order swapped and the target page rendered as structured text or prose. A structured rendering "raises target citation count by +0.50 citations per answer" but did so "without increasing total citations or reducing competitor credit"; the effect on whether the page is cited at all was +4.5 points and "inconclusive". Pages in position 1 were cited 85.1% of the time against 42.8% in position 5, but a controlled reorder moved only 7.9 points. On fresh reruns, "15 percent of binary decisions change." Limits: one model and one search provider, replayed transcripts rather than live pages, preprint.

Method: 10,000 questions from 20 advice and support subreddits, each run three times about five hours apart through Reddit's own AI search, 30,000 answers traced back through 14.68 million comments. Comments one standard deviation more formal had "49% higher odds of being selected"; the median selected comment sat at the 91st percentile for score against the 45th for comments not selected; 92% of selected comments were direct replies to the original post; first-person language fell from 3.3% in quoted comments to 0.06% in the final answers. Limits: preprint, Reddit's own AI search only, and the authors say it "may not carry over" to other AI search systems.