Clocking AI Citations

To this point, AEO (Answer Engine Optimization) has felt like optimizing into the fog.

SEO at least provides familiar signals: rankings, impressions, clicks, crawl status, and Search Console. With these new answer engine LLMs, it gets messier. You publish a page, and maybe ChatGPT, Claude, etc., cite it somewhere at some point (insert me waving my hands and arms around in the air). Very normal, very chill, definitely not maddening.

That’s why the LinkedIn post below (and the associated graph) from Josh Blyskal of the founding team at Profound caught my attention:

At first glance, the data and conclusions feel like a big deal. AEO may finally have a clock!

But once you look closer, the better question becomes: what exactly is that clock measuring? Objectively, the chart is interesting, but the post is making the data carry more weight than it probably can.

The biggest issue is that it treats “time to first citation” like a universal AEO benchmark, when it is really a benchmark for a specific observed dataset: ~900 newly published marketing pages that were eventually cited by ChatGPT or Claude agents in Profound’s logs during a specific 60-day window. That’s useful, but it isn’t the same thing as saying, “90% of new pages will get cited by ChatGPT or Claude within 37 days.”

The basic metric may be directionally useful. If Profound is tracking real agent/log behavior across ChatGPT and Claude, then the median, P75, and P90 numbers are valid for the pages in that sample. The post claims the dataset was about ~900 newly published marketing pages, and the graphic frames it as “new pages cited by ChatGPT or Claude agents” from March to May 2026. That phrasing matters a lot.

The post is also correct that crawlability matters. OpenAI has separate crawler/user-agent behavior: OAI-SearchBot is used to surface websites in ChatGPT search answers, GPTBot is for training-related crawling, and ChatGPT-User may visit pages during user-initiated actions. OpenAI also says sites that opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they may still appear as navigational links.  

It’s also fair to say Claude has separate crawler behaviors. Anthropic documents ClaudeBot for model development, Claude-User for user-directed retrieval, and Claude-SearchBot for improving search result quality. Anthropic specifically says disabling Claude-SearchBot may reduce a site’s visibility and accuracy in user search results.  

So yes: if your robots.txt, CDN/firewall, rendering, canonicalization, or page architecture prevents these systems from accessing a page, that can absolutely hurt visibility. That part is not SEO snake oil, it’s plumbing.

The line “by 37 days, 90% of pages have gotten their first citation” is providing the biggest challenge for me. More accurately, it should say:

“Among the ~900 newly published marketing pages in this dataset that were cited at least once, 90% received their first observed citation within 37.1 days.”

That is a very different claim.

Is there selection bias baked in? The chart is about pages that were cited, not all pages published. It doesn’t tell us what percentage of newly published pages never got cited at all, and that missing denominator is the whole ballgame. Saying “if you’re beyond 37 days without citations, the issue is most likely on your end” goes too far.

Maybe…

Also, “ChatGPT or Claude” is too blended. Those aren’t one engine, obviously. Even within ChatGPT, source behavior differs depending on whether the answer uses search, deep research, agentic browsing, connectors, or no web retrieval at all. OpenAI says ChatGPT will automatically search when a question might benefit from web information, and responses that use search may include citations. That means many ChatGPT responses are not citation events, and many citation events are query-dependent, not purely index-dependent.

The post also seemingly conflates being accessible/indexed with being cited. A page can be crawlable and still never be cited because citations are chosen in context. The retrieval layer is trying to satisfy a user’s specific question. Citation is not a reward for publishing. It is more like being picked from the lineup when the matchup makes sense.

The “under 6 days means you’re doing something right” claim is partially true but incomplete. Fast citation could mean good technical access, strong brand/domain authority, timely relevance, good internal linking, or newsworthiness. It could also mean the page was created in a category where Profound’s monitored prompts were already asking related questions. Not exactly a victory parade, more like “your page got on the field early.”

For ChatGPT, there is no single universal “citation clock.” It doesn’t just absorb a new page into its base model knowledge a few days after it is published. When it cites a new page, it usually happens because a live retrieval/search system finds it during a user request, or because a tool/browser/search layer has access to it.

That means a fresh page can be cited quickly IF it is discoverable through search/retrieval and relevant to the prompt. But that does not mean the model “knows” the page in the way we as marketers often imagine. There is a meaningful difference between:

“ChatGPT can retrieve and cite this page during search.”
and
“ChatGPT has internalized this page into its model knowledge.”

The post mostly talks about the first one, but some marketers will read it as the second, and that’s where things get messy.

Based on Anthropic’s documentation, Claude has similar but distinct moving parts: training crawler, search/indexing crawler, and user-directed retrieval agent. So the same warning applies: “Claude cited this page” does not necessarily mean the base model has learned it. It may mean Claude’s search/retrieval path found it for a specific user query.

The data is useful as an early benchmark for observed first-citation timing among already-cited pages, but the post becomes shaky when it turns that into an AEO law of physics.