How to check whether AI answers cite your site.

Why a method, not a glance

Generated answers are not stable. Atil et al. ran five language models, configured to be deterministic, ten times on the same benchmark inputs. None delivered repeatable accuracy across all tasks, and accuracy varied by up to 15% between runs [1]. Vendors that track AI answers say the same about sources. One vendor’s documentation notes that a source “might appear one day and be absent the next” [2]. Another says the same prompt “can return different results on different runs” [3]. If you check once and see your site, you have learned that it can appear. You have not learned how often it does.

Citations also need reading, not only counting. In an audit of four generative search engines, 74.5% of citations supported the sentence they were attached to [4]. A cited URL can sit next to a claim your page never made.

The method

1. Write the question set. List the questions a buyer would type, in their words, before they know your brand. Keep branded questions (that name you) in a separate group. Give each question an ID and freeze the wording. A changed word is a new question.
2. Fix the conditions. For each AI product, write down the mode (for example, with or without web search), whether you are logged in, your language and your location. One vendor notes that a logged-out session lets the platform decide which model to use and whether to search the web [5]. Use the same setup every round. Start a new chat for every run, so earlier turns do not shape the answer.
3. Decide the number of runs before you start. Choose how many times each question runs per round, and spread runs across at least two days. Keep that number fixed for every round you compare. Deciding afterwards invites you to stop when the result looks good.
4. Record every run in one row. Use the columns below.
5. Summarise per question and per product. Report three plain rates: the share of runs that cite your domain, the share that name your brand, and, among your citations, the share where the sentence is fully supported. Show the number of runs next to every rate. A rate from few runs should say so.
6. Check the other side in your own data. In your server logs, look for the AI user agents. OpenAI documents OAI-SearchBot for search and ChatGPT-User for user-triggered fetches [6]. Anthropic documents Claude-SearchBot and Claude-User [7]. Perplexity documents PerplexityBot and Perplexity-User [8]. For Google, traffic from AI Overviews and AI Mode is counted inside Search Console’s “Web” search type [9]. These logs tell you whether your pages are fetched. The answer sheet tells you whether they are cited.
7. Repeat on a schedule. Run the same set on the same conditions each round. Compare round to round only when nothing in the first three steps changed.

What this method does not tell you

• Why an engine cited a source. C-SEO Bench found most conversational SEO methods largely ineffective [10], so be careful when reading any change as the effect of your edit.
• How many people ask a question. The sheet records answers, not demand.
• Other people’s answers. Personal settings, location and history can differ from yours. Your sheet describes your conditions.

One row from our own data

Here is how the sheet links to a real measurement. On our own domains, the Axis dashboard, read on 8 October 2026, shows 0 sessions from AI assistants, with “19 of 28 days measured” [11]. On its own, that number cannot tell you whether the sites are never cited or are cited and not clicked. The answer sheet is how you tell those two cases apart.

Map

Look. Freeze the questions and the conditions.

Find. Record every source, every run, and read the cited sentence.

Align. Improve the pages that should have been cited, then run the same sheet again.

References

1. Atil, B., Aykent, S., Chittams, A., Fu, L., Passonneau, R. J., Radcliffe, E., Rajagopal, G. R., Sloan, A., Tudrej, T., Ture, F., Wu, Z., Xu, L., & Baldwin, B. (2024). Non-determinism of “deterministic” LLM settings. arXiv:2408.04667. https://arxiv.org/abs/2408.04667
2. Peec AI. (n.d.). Understanding chats. Peec AI Docs. Accessed 9 October 2026. https://docs.peec.ai/understanding-chats
3. OtterlyAI. (n.d.). How often does OtterlyAI check AI search engines? Help Center. Accessed 9 October 2026. https://help.otterly.ai/monitoring-interval
4. Liu, N. F., Zhang, T., & Liang, P. (2023). Evaluating verifiability in generative search engines. Findings of EMNLP 2023. https://aclanthology.org/2023.findings-emnlp.467/
5. Peec AI. (n.d.). Welcome to Peec AI. Peec AI Docs. Accessed 9 October 2026. https://docs.peec.ai/intro-to-peec-ai
6. OpenAI. (n.d.). Overview of OpenAI crawlers. Accessed 9 October 2026. https://platform.openai.com/docs/bots
7. Anthropic. (2026, April 7). Does Anthropic crawl data from the web, and how can site owners block the crawler? Claude Help Center. Accessed 9 October 2026. https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler
8. Perplexity. (n.d.). Perplexity crawlers. Accessed 9 October 2026. https://docs.perplexity.ai/guides/bots
9. Google Search Central. (n.d.). AI features and your website. Accessed 9 October 2026. https://developers.google.com/search/docs/appearance/ai-features
10. Puerto, H., Gubri, M., Green, T., Oh, S. J., & Yun, S. (2025). C-SEO Bench: Does conversational SEO work? NeurIPS 2025 Datasets and Benchmarks Track. https://arxiv.org/abs/2506.11097
11. Axis. (2026, October 8). Dashboard reading for Axis’s own domains, 28-day window. Internal measurement, not a customer result.

If you ran your main buyer questions tomorrow, which domains would you expect to see cited? Tell us where you want to stand.