agentlinkopsOne backlink question, eight AI-cited sources
What one dated Claude answer reveals about the pages selected for a backlink question, and why one observation cannot establish AI citation share.
One answer to a backlink question cited eight pages. None was ours. The record shows which sources one assistant selected. It cannot establish our share of AI citations or predict its next answer.
One observation, made on the day this was written
On September 15, 2026 at 16:59 UTC, we asked Claude Code with web search this exact question:
What should I check before concluding that a backlink was removed from a web page?
The answer cited eight pages. The figure numbers them in the order the answer listed them and links each one.
The eight cited pages, grouped by kind
2 Supplier help articles
1 Competitor article
5 Third-party posts
0 Our own site
- agentlinkops.com
Two are supplier help articles. One is an article by LinkGuard, a monitor we cover in our backlink monitor comparison. Five are third-party posts. The record measures one answer from one model on one date, with no own-site citation.
What the source list tells us
The question asks for evidence before declaring a link removed. Several selected pages address lost-link reasons, false alerts or JavaScript. That gives us specific pages and explanations to study, rather than a broad instruction to publish more content.
A page in the list is a research lead. A separate page-link check shows whether it links to us, and the publisher's own guidelines show whether it accepts contributions. The assistant's selection and the publisher's editorial decision remain separate.
What the platforms say about their own citations
These are published statements checked on September 15, 2026, not platform tests we ran. OpenAI distinguishes its search crawler from user-triggered page visits. Anthropic's web-search documentation says citations carry source information; its crawler guidance separately describes website access. Perplexity's API quickstart describes answers with inline citations.
Google says its AI search features require no additional special markup beyond its usual search requirements. It includes their traffic in overall Search Console traffic. Its crawler documentation also distinguishes Google-Extended controls from inclusion in Google Search.
These statements help interpret logs, but a crawler request cannot establish that a particular answer cited a page. The readable answer is the evidence for that observation.
Measuring answers carefully
A blocked browser or unreadable source list is an unknown attempt, not an answer with zero citations. If only one of several attempts succeeds, reporting the successful answer alone conceals a measurement limit.
Answers can change with the query, model, surface and date. Two URLs on the same host remain two source pages. JavaScript can also change what a browser shows: a complete static response can lack a link that appears after interaction. Keep that browser observation separate from the fetch result.
To estimate how often a page is cited, ask the same question several times and report the count with the number of attempts. The citation run reference describes the panel format and verdicts.
What to do with a cited page
Before you write about a question an assistant already answers, read the pages it cites. Look for an explanation that is missing or weak, and publish only when your evidence improves on it. Another version of the same checklist gives a reader nothing new.
To turn cited pages into outreach targets, see link-building campaigns.
Sources (7)
- Overview of OpenAI crawlers · OpenAI · 2026-09-15
- Does Anthropic crawl data from the web, and how can site owners block the crawler? · Anthropic · 2026-04-07
- Web search tool · Anthropic · 2026-09-15
- AI features and your website · Google Search Central · 2025-12-10
- List of Google's common crawlers · Google Search Central · 2026-07-14
- Agent API quickstart · Perplexity · 2026-09-15
- JavaScript SEO basics · Google Search Central · 2026-03-04