To get cited by ChatGPT, Perplexity or Google AI Overviews, let their search crawlers in (OAI-SearchBot, PerplexityBot, Googlebot), open each page with a 40 to 60 word answer, name a real author, date every figure and link it to its source. Google says no special markup is required for AI Overviews (Search Central, December 2025).
Key takeaways
- Search and training are separate crawlers. OAI-SearchBot feeds ChatGPT search; GPTBot feeds training. Blocking the second does not touch the first. OpenAI developer documentation (2 September 2026)
- Google needs nothing special. "There are no additional requirements to appear in AI Overviews or AI Mode" beyond being indexed with a snippet. Google Search Central (10 December 2025)
- Citations, quotations and statistics raised visibility in generative responses by up to 40 percent in the paper that defined GEO. arXiv 2311.09735 (v3, 28 June 2024)
- The click is shrinking. 8 percent of Google visits with an AI summary ended in a click on a result, 15 percent without. Pew Research Center (22 July 2025)
- The audience is real. 7 percent use an AI chatbot for news each week, 15 percent of under-25s; 47 percent of journalists use ChatGPT. Reuters Institute for the Study of Journalism (17 June 2025) Muck Rack (19 March 2026)
How ChatGPT, Perplexity and Google decide what to cite
Each engine runs a search crawler that is distinct from its training crawler, and each says so on its own documentation page. OpenAI: OAI-SearchBot is "used to surface websites in search results in ChatGPT's search features", GPTBot crawls content "that may be used in training", and ChatGPT-User, which fetches a page when a user asks, is "not used to determine whether content may appear in Search" (OpenAI, crawlers). Perplexity: PerplexityBot exists to "surface and link websites in search results on Perplexity" and is "not used to crawl content for AI foundation models" (Perplexity, crawlers). Anthropic: Claude-SearchBot "navigates the web to improve search result quality", ClaudeBot collects content for training (Anthropic, crawlers).
Google is the exception: AI Overviews and AI Mode draw from the ordinary index. A page needs to be "indexed and eligible to be shown in Google Search with a snippet", and "there's also no special schema.org structured data that you need to add" (Google, AI features and your website, updated 10 December 2025). Google-Extended, the token people block to opt out of Gemini training, "does not impact a site's inclusion in Google Search nor is it used as a ranking signal" (Google, common crawlers, updated 14 July 2026).
After retrieval, every engine does the same thing: it extracts a passage, blends it with passages from other sources, and links back. So the work is in two places, access and the passage. Steps 1 to 7 cover both.
Step 1: Let the search crawlers in
Check robots.txt before touching a word of copy. The table lists the search agent per operator and the training agent you may block without losing citations.
| Allow this agent | What it feeds | Training agent you may block instead |
|---|---|---|
OAI-SearchBot | ChatGPT search results. OpenAI developer documentation (2 September 2026) | GPTBot (training) |
PerplexityBot | Perplexity search results and links. Perplexity documentation (2 September 2026) | none documented for training |
Claude-SearchBot | Claude web search results. Anthropic support (2 September 2026) | ClaudeBot (training) |
Googlebot | Google Search, AI Overviews and AI Mode. Google Search Central (14 July 2026) | Google-Extended (Gemini training) |
A minimal block that keeps citations open while opting out of one training crawler:
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Claude-SearchBot
Allow: /
# Training crawlers, your choice; blocking them does not remove citations
User-agent: GPTBot
Disallow: /Then read the server logs a week later. If OAI-SearchBot and PerplexityBot never appear, the page is not reachable or not discoverable, and no copy change will fix that.
Step 2: Answer in the first 50 words
The engine cites a passage. Give it one: a paragraph of 40 to 60 words directly under the H1 that answers the question the page title asks, with the key number and its source in parentheses. The pages that answer engines cited in our September 2026 audit of press release queries opened with blocks of 47 to 71 words; ours ran from 57 to 113. Shorter won.
Write the definition as a sentence that can stand alone: "A press release is an official statement issued to the media to announce something newsworthy." Then the number. Then the source. Marketing language in the first paragraph sends the engine to the next result.
Step 3: Put a real name and a date on the page
Google's Article documentation lists author, author.name, author.url, datePublished, dateModified, headline and image as recommended properties, and asks that "all the authors that are presented as authors on the web page are also included in markup" (Google, Article structured data). That rules out the common setup: "editorial team" on the page and a named person in the JSON-LD. Pick the person, show the name, the role and the update date under the H1, link to a profile page, and declare the same Person once, by @id, rather than redeclaring it on every page.
Step 4: Date every figure and link it to its source
The paper that coined Generative Engine Optimization, presented at KDD 2024, tested which page changes moved visibility in generative responses and found gains of "up to 40%" from adding citations, quotations and statistics (Aggarwal et al., arXiv 2311.09735). A figure with a date and a link is a fact the engine can verify by following the link; a figure without one is a claim it can drop. Put the month and year in the sentence, and make the number the link, not the word "source".
Step 5: Use lists and tables with stable columns
A numbered procedure is cited item by item. A comparison table is cited row by row. A paragraph is paraphrased, and the paraphrase may drop your name. Keep the same column names on every page of the same type, so the extraction is predictable: Plan, Price, Contract, Source. Put the source inside the cell, with its date.
Step 6: Keep one topic per page and a clean heading tree
One page answers one question. H1, then H2, then H3, no skipped level, H2s phrased as the question a person types. A long guide that answers six questions competes with itself and gives the engine six weak passages instead of one strong one. Split it into a hub and six spokes, each with its own answer block.
Step 7: Re-check quarterly and log the citations
Measurement is a manual probe. Run your top 10 buyer questions in ChatGPT with search, Perplexity, Claude with web search and Gemini every week; run the full list of 50 to 100 monthly. Log the domain cited, the passage used and whether your brand was named. Add referral traffic from the assistants in your analytics. When a fact on a page changes, update the copy and dateModified rather than publishing a second page on the same topic.
What the data says
- 7 percent of people across 48 markets use an AI chatbot for news each week, 15 percent of under-25s, with ChatGPT and Gemini named as the examples. Reuters Institute for the Study of Journalism (17 June 2025)
- 18 percent of Google searches in a March 2025 panel of 900 US adults produced an AI summary; a result was clicked in 8 percent of those visits against 15 percent without a summary, and a link inside the summary in 1 percent. Pew Research Center (22 July 2025)
- 82 percent of journalists use at least one AI tool, up from 77 percent; 47 percent use ChatGPT and 22 percent Gemini, in a survey of 897 journalists fielded January 30 to March 2, 2026. Muck Rack (19 March 2026)
- Up to 40 percent visibility gain in generative engine responses from citations, quotations and statistics, on a benchmark of diverse queries. arXiv 2311.09735 (v3, 28 June 2024)
Mistakes that hurt your chances
- Blocking the wrong bot. A robots.txt that disallows OAI-SearchBot to stop training blocks citations and leaves training untouched, because GPTBot is the training crawler.
- Burying the answer. A first paragraph of positioning sends the engine to the next source. Start with the fact.
- Two authors, or none. A team byline on the page and a person in the markup is the mismatch Google's Article documentation warns against.
- Numbers without links. A statistic the engine cannot verify is a claim it can drop, and the paper that defined GEO found sourced statistics to be the lever.
- Client-side JSON-LD. Markup injected by JavaScript is invisible to a crawler that does not execute scripts. Server-render it.
- Stale dates. A page last updated in April with September prices reads as abandoned to a person and to an engine. Update the copy and the date together.
Press releases: the citation asset most companies waste
A press release already has what an engine wants: named entities, a date, numbers, a quote and a publisher. What it usually lacks is the publishing layer: an author, a canonical newsroom page, dated figures with links, and open crawlers. The press release GEO guide covers the 12 writing patterns and the schema stack; the press release SEO guide covers the canonical and syndication rules. PressPilot hosts each release on a newsroom page with Article markup and a named author, and sends it to journalists you pick from 45,476 contacts (live count, September 2026).
Frequently asked questions
- Does ChatGPT cite sources?
- Yes, in its search features. OpenAI documents OAI-SearchBot as the crawler "used to surface websites in search results in ChatGPT's search features", and ChatGPT-User as the agent that fetches a page when a user asks. A citation is never guaranteed: the page has to be crawlable, answer the question in one passage, and carry an author and a date.
- Do I need to allow GPTBot to be cited by ChatGPT?
- No. GPTBot crawls content "that may be used in training" OpenAI's models. OAI-SearchBot is the one that feeds ChatGPT search. You can block GPTBot and keep OAI-SearchBot allowed, and the two decisions are independent. The same split exists at Anthropic (ClaudeBot for training, Claude-SearchBot for search) and at Google (Google-Extended for Gemini training, Googlebot for Search).
- How can I see whether ChatGPT cites my site?
- Ask it. Run your 10 most important buyer questions in ChatGPT with search enabled, weekly, and record which domains are cited and which passage was used. Add referral traffic from chatgpt.com and perplexity.ai in your analytics. We have not found an official citation report from OpenAI, so the manual probe is the measurement.
- Do ChatGPT citations help Google SEO?
- Not directly, and we know of no Google statement that they do. What they bring is referral traffic and brand recall in a place where clicks are scarce: Pew measured a click on a result in 8 percent of Google visits with an AI summary against 15 percent without. The same page structure serves both, so there is no trade-off to make.
- What is the difference between ChatGPT search and Google AI Overviews?
- Different crawlers, same needs. ChatGPT search uses OAI-SearchBot and its own retrieval. Google AI Overviews use the normal Google index: Google says a page needs to be "indexed and eligible to be shown in Google Search with a snippet" and that no special optimization or schema is required. Both cite short passages from pages they can read.
- Does schema markup get a page cited by AI?
- It helps the engine attribute the page, not select it. Google states there is "no special schema.org structured data" to add for AI Overviews, and separately lists author, author.url, datePublished and dateModified as recommended Article properties. Use markup to declare one author and real dates. Do not expect it to substitute for a clear answer.
- How long does it take to get cited by ChatGPT?
- As long as it takes the search crawler to fetch and index the page, which no operator publishes as a number. Check your server logs for OAI-SearchBot and PerplexityBot after publishing, then probe weekly. If the bots never visit, the problem is access or discoverability, not content.
- Should I optimize for ChatGPT or for Google?
- Both, on the same page. Among journalists, ChatGPT is used by 47 percent and Gemini by 22 percent (Muck Rack, 2026). Among the public, 7 percent use an AI chatbot for news weekly, 15 percent of under-25s (Reuters Institute, 2025). Google still carries most visits. One page with an answer block, an author, dated sources and open crawlers serves all of them.
Related reading
- Press release GEO: 12 writing patterns that get cited
- Press release SEO in 2026
- What is an AI Overview (glossary)
- What is GEO (glossary)
- Press release distribution guide
Publish releases that AI engines can attribute
PressPilot writes the release with the answer block, hosts it on a newsroom page with Article markup, one named author and visible dates, and sends it to the journalists you pick. From 29 EUR a month for 100 journalist credits, no contract.