https://unsourced.app/research Short, sourced takes on AI-search news and original field research from Unsourced. One new post every Tuesday. en-gb Tue, 03 Nov 2026 08:00:00 GMT https://unsourced.app/research/identity-or-path-two-ways-to-clean-ai-crawler-logs https://unsourced.app/research/identity-or-path-two-ways-to-clean-ai-crawler-logs Tue, 03 Nov 2026 08:00:00 GMT 6,805 AI-crawler visits checked two ways. 83% of unverified visits asked for pages that never existed. 342 hit real pages; only the identity check caught them. https://unsourced.app/research/seo-agencies-least-cited-by-ai https://unsourced.app/research/seo-agencies-least-cited-by-ai Tue, 27 Oct 2026 08:00:00 GMT We checked 180 UK agencies across five AI models. SEO firms were named least (2 of 58), PR firms most (9 of 56). For SEO questions, AI cited Google. https://unsourced.app/research/ai-search-citations-are-a-long-tail https://unsourced.app/research/ai-search-citations-are-a-long-tail Tue, 20 Oct 2026 08:00:00 GMT 100 questions, 4 AI engines, 1,032 cited domains. The top site got 2.4% of citations and 69% of domains were cited once. There is no top 10 to join. https://unsourced.app/research/ai-citation-leaderboards-measure-the-wrong-thing https://unsourced.app/research/ai-citation-leaderboards-measure-the-wrong-thing Tue, 13 Oct 2026 08:00:00 GMT AI engines overlap only about 8% on cited domains, and half of cited sources change on a re-ask. Why one leaderboard score cannot hold, from our data. https://unsourced.app/research/most-agencies-are-not-cited-for-their-own-topics https://unsourced.app/research/most-agencies-are-not-cited-for-their-own-topics Tue, 06 Oct 2026 08:00:00 GMT A quick AI citation check on 52 UK marketing and SEO agencies: 85% were not named for their own topics. AI named a median of 35 other sites instead. https://unsourced.app/research/ai-cites-different-sources-each-time https://unsourced.app/research/ai-cites-different-sources-each-time Tue, 29 Sep 2026 08:00:00 GMT We asked four AI search engines the same questions three times. About half the cited sources changed on an identical re-ask. One check is not proof. https://unsourced.app/research/gemini-cites-google-youtube https://unsourced.app/research/gemini-cites-google-youtube Sat, 26 Sep 2026 08:00:00 GMT Across 100 questions, Gemini sent 1 in 15 citations to YouTube. ChatGPT, Copilot and Perplexity sent almost none. Engine owners shape their sources. https://unsourced.app/research/chatgpt-cites-wikipedia-perplexity-cites-reddit https://unsourced.app/research/chatgpt-cites-wikipedia-perplexity-cites-reddit Tue, 15 Sep 2026 08:00:00 GMT The big AI engines cite different corners of the web: ChatGPT leans on Wikipedia, Perplexity on Reddit, and only about 11% of cited domains overlap. https://unsourced.app/research/llms-txt-wont-get-you-cited https://unsourced.app/research/llms-txt-wont-get-you-cited Tue, 08 Sep 2026 08:00:00 GMT Server logs show the AI labs llms.txt is meant for fetch it roughly never, and there is no measurable citation lift. Optimise for evidence instead. https://unsourced.app/research/blocking-ai-crawlers-is-not-one-switch https://unsourced.app/research/blocking-ai-crawlers-is-not-one-switch Tue, 01 Sep 2026 08:00:00 GMT Cloudflare's default AI block splits crawlers into Training, Search and Agent and blocks only some, with labels the AI companies write. What it does. https://unsourced.app/research/verify-before-you-block-ai-crawlers https://unsourced.app/research/verify-before-you-block-ai-crawlers Tue, 25 Aug 2026 08:00:00 GMT Block, charge or allow is now a decision every site faces. But the user-agent you'd base it on is a claim, not proof. How to verify a crawler first. https://unsourced.app/research/one-in-five-ai-crawlers-are-impostors https://unsourced.app/research/one-in-five-ai-crawlers-are-impostors Thu, 20 Aug 2026 08:00:00 GMT Over 30 days, 21.8% of requests with an AI crawler's user-agent were not that crawler. The same IPs claimed five AI companies at once. The data inside. https://unsourced.app/research/ai-engines-cite-different-sources https://unsourced.app/research/ai-engines-cite-different-sources Tue, 11 Aug 2026 08:00:00 GMT We asked Gemini, Perplexity and ChatGPT the same 156 questions. They cited the same domains about 8% of the time. 91% of each engine's sources were its own. https://unsourced.app/research/how-ai-visibility-scores-work https://unsourced.app/research/how-ai-visibility-scores-work Tue, 11 Aug 2026 08:00:00 GMT How AI visibility scores are built, and why one number can't be right when three AI engines agree on a source only about 8% of the time. https://unsourced.app/research/ai-crawlers-see-a-different-web https://unsourced.app/research/ai-crawlers-see-a-different-web Mon, 10 Aug 2026 08:00:00 GMT Time serves approved AI crawlers a machine-readable site, with sponsored answers readers never see. What that means and how to check your own site. https://unsourced.app/research/crawl-is-not-a-citation https://unsourced.app/research/crawl-is-not-a-citation Tue, 04 Aug 2026 08:00:00 GMT Cloudflare's Pay Per Use pays publishers when an AI answer uses their content, not when a bot fetches it. Why a crawl and a citation are different. https://unsourced.app/research/eu-ai-act-compliance-verification-gap https://unsourced.app/research/eu-ai-act-compliance-verification-gap Tue, 28 Jul 2026 08:00:00 GMT From 2 August 2026 the EU enforces AI transparency rules, including copyright opt-outs. Nothing in the mandatory template lets outsiders verify them. https://unsourced.app/research/impression-citation-referral https://unsourced.app/research/impression-citation-referral Wed, 22 Jul 2026 08:00:00 GMT AI search visibility is three measurements: whether AI showed you, named you as a source, or sent you a visitor. What each means and who measures it. https://unsourced.app/research/ai-engines-pay-cited-publishers https://unsourced.app/research/ai-engines-pay-cited-publishers Mon, 13 Jul 2026 08:00:00 GMT Perplexity's Comet Plus reportedly puts $42.5M behind the publishers it cites, 80/20 in their favour. When citations pay, proving you were cited matters. https://unsourced.app/research/robots-txt-cant-stop-ai-scrapers https://unsourced.app/research/robots-txt-cant-stop-ai-scrapers Tue, 07 Jul 2026 08:00:00 GMT Scrapers wear a Googlebot name tag and walk through the door you leave open for the real one. Why a bot's self-reported name won't do, and what will. https://unsourced.app/research/bing-ai-performance-4-things-it-wont-tell-you https://unsourced.app/research/bing-ai-performance-4-things-it-wont-tell-you Thu, 02 Jul 2026 08:00:00 GMT Bing Webmaster Tools now reports AI citations, Intents, Topics and Citation Share. A real milestone, but single-engine. Here is what it leaves out. https://unsourced.app/research/ai-visibility-score-is-a-guess https://unsourced.app/research/ai-visibility-score-is-a-guess Tue, 30 Jun 2026 08:00:00 GMT Most AI-visibility tools give you a number out of 100. Is it measured or modelled? A test you can run on any tool, including ours, to tell which. https://unsourced.app/research/forensic-geo-agency-retention https://unsourced.app/research/forensic-geo-agency-retention Tue, 23 Jun 2026 08:00:00 GMT Rankings can hold while AI citations collapse, and SEO reports miss it. How to swap a vanity score for an evidence-led audit that anchors a retainer. https://unsourced.app/research/ai-search-stop-optimising-start-auditing https://unsourced.app/research/ai-search-stop-optimising-start-auditing Tue, 23 Jun 2026 08:00:00 GMT Your content is now the AI supply chain, not a traffic source. The fix is auditing who extracts your data and what you get back, not more SEO. https://unsourced.app/research/bait-keep-or-block-ai-crawlers https://unsourced.app/research/bait-keep-or-block-ai-crawlers Tue, 16 Jun 2026 08:00:00 GMT Blocking every AI bot erases you from answers you want to win. There are three kinds of crawler: bait the search bots, keep the trainers, block the rest. https://unsourced.app/research/ai-crawler-spoofing https://unsourced.app/research/ai-crawler-spoofing Tue, 09 Jun 2026 08:00:00 GMT Your logs say GPTBot, but that name is a header the visitor writes itself. How to verify an AI crawler at the network layer, with three honest verdicts.