Block Gemini and It Stops Citing You. Google's AI Mode Cites the Same Outlets in 69% of News Answers.
Most big newsrooms now block some AI engines and not others. Until now there has been little hard evidence of what those blocks actually do to who gets cited. A new study from Five Blocks, a reputation management firm, puts numbers on it.
It read the robots.txt files and page tags of 44 major news outlets on 23 October 2026, then asked eight AI engines six questions a day about every Fortune 100 company and its CEO from 24 to 26 October, collecting 14,392 answers (vendor-stated, Five Blocks). The result is the clearest picture yet of blocking as it really works: selective, leaky, and controlled more by the engine than by the publisher.
Which AI engines do news outlets block?
Of the 44 outlets, 25 block Google's Gemini app, 25 block Perplexity's crawler, 18 block Claude, 14 use a page tag that keeps them out of Copilot, 12 block ChatGPT's search bot and 4 block Grok (vendor-stated, Five Blocks). None block Google's AI Overviews or AI Mode.
That last number is structural, not a choice. AI Overviews and AI Mode are built on Googlebot, Google's ordinary search crawler. Blocking it means leaving Google Search. Google's own documentation, quoted in the report, says the Google-Extended token that controls Gemini "does not impact a site's inclusion in Google Search".
Twelve outlets block no engine at all. At the other end, the New York Times, the Wall Street Journal, Barron's, MarketWatch, the Economist and NPR leave only three of the eight engines unblocked (vendor-stated, Five Blocks).
Do the blocks hold?
Five Blocks compared how often each engine cited outlets that block it against how often it would have done if the block made no difference. The answers across four news questions over three days:
- Gemini app: 0 seen against 348 expected.
- ChatGPT: 28 against about 338, rising each day from 1 to 7 to 20.
- Copilot: 29 against 133.
- Claude: 99 against 306.
- Perplexity: 124 against 172.
- Grok: 312 against 228.
(All vendor-stated, Five Blocks.) The Gemini app, ChatGPT and Copilot largely respect blocks. Claude cites blocked outlets about a third as often as expected. Perplexity, whose live fetcher it says "generally ignores robots.txt rules", cites them anyway. Grok publishes no crawler name, so it is close to unblockable, and it cited blocked outlets more often than expected.
What is the Google split?
This is the finding that matters most. The 25 outlets that block the Gemini app were cited in zero Gemini app answers, and in 69% of Google AI Mode news answers (vendor-stated, Five Blocks). Yahoo Finance appeared in 46.4% of AI Mode news answers and none from the Gemini app; Reuters in 22.3% against none.
Same company, two AI products, two rulebooks. An outlet that blocks Gemini has not kept its journalism out of Google's AI. It has kept its name out of one Google product, while the larger one keeps reading and quoting it.
What else did the study find?
Deals buy citations on ChatGPT. ChatGPT cited the 12 outlets with OpenAI content deals in 44.2% of news answers, against 23.3% for other open outlets. AI Mode, with no such deals, showed almost no gap (vendor-stated, Five Blocks).
Paywalls rarely stop crawlers. Of five paywalled sites Five Blocks could test, four sent the full article in the page and hid it from readers with a script. The Wall Street Journal, behind a hard paywall, appeared in 5.1% of AI Mode news answers.
Stock sites fill the gap. Sites such as MarketBeat and Investing.com appeared in 40% to 60% of news answers on every engine. The Gemini app's single most-cited domain was marketbeat.com, at 21%.
Answers churn. Only 30% to 60% of links in a news answer came back the next day for the same question (vendor-stated, Five Blocks).
The limits are real: Fortune 100 companies only, six questions, three days, and only the links an engine chose to show. As Five Blocks says, nobody outside the AI companies can see what an engine read but did not link.
What does this mean for publishers?
Blocking is not a single lever. It is eight different levers, and the publisher controls only some of them. Against Gemini, ChatGPT and Copilot a block mostly removes you from the answer. Against Perplexity and Grok it mostly does not. Against AI Overviews and AI Mode it is not available without leaving search.
So the practical effect of blocking is selective invisibility. Your reporting keeps informing the answer through engines you cannot block, through syndicated copies, and through stock sites rewriting your story, while the credit flows to whoever was left open. The study's own conclusion for communicators is to add one column to every media list: which engines can use each outlet.
What does this mean for advertisers?
For brands, the source list behind an AI answer is now an engine-by-engine question. A Wall Street Journal story about your company will rarely surface as a Copilot or Gemini link; a MarketBeat page summarising it may. Brand visibility in LLM answers depends on which publishers each engine can read and chooses to show, and that set moves daily. One screenshot of an answer tells you very little.
the blankspace read: This study is the best evidence yet that blocking is a trap. It does not stop the read: Google's AI Mode reads the same pages through a crawler no publisher can refuse, and Perplexity and Grok read around the rules. What blocking removes is the credit, and with it any route to being paid. The reads that keep arriving are Live Search Agents fetching pages in real time, running no JavaScript, loading no ad tags, invisible to analytics. The question is not whether to let them in. It is whether every read that gets through earns anything. Monetise the read, not the referral.
Key figures
| Figure | Value | Provenance |
|---|---|---|
| Outlets studied | 44 major news outlets | Vendor-stated (Five Blocks) |
| AI answers analysed | 14,392 (8 engines, 24 to 26 Oct 2026) | Vendor-stated (Five Blocks) |
| Outlets blocking the Gemini app | 25 of 44 | Vendor-stated (Five Blocks) |
| Outlets blocking ChatGPT's search bot | 12 of 44 | Vendor-stated (Five Blocks) |
| Outlets blocking AI Overviews or AI Mode | 0 of 44 | Vendor-stated (Five Blocks) |
| Gemini app citations of outlets blocking it | 0 seen vs 348 expected | Vendor-stated (Five Blocks) |
| AI Mode news answers citing those same outlets | 69% | Vendor-stated (Five Blocks) |
| Grok citations of outlets blocking it | 312 seen vs 228 expected | Vendor-stated (Five Blocks) |
| ChatGPT news answers citing OpenAI deal outlets vs other open outlets | 44.2% vs 23.3% | Vendor-stated (Five Blocks) |
| News answers citing stock or investing sites | 40% to 60%, every engine | Vendor-stated (Five Blocks) |
| Links back the next day, news questions | 30% to 60% | Vendor-stated (Five Blocks) |
Primary source, opened and verified: Which News Sites Can AI Read, and Which Does It Cite? Five Blocks, 5 November 2026 (full report PDF linked on page)