How often should I re-check AI visibility, and what changes between checks?

Nobody has measured an optimal re-check interval. Google, OpenAI, Anthropic and Perplexity publish no guidance on it, and no study in our evidence base tests one. A weekly or monthly rhythm is fine for reporting, but it is a convention.

What matters more is knowing what can move between two checks. There are four sources of change:

1. The model and its routing. The vendor swaps or updates the model behind the app.
2. The retrieval decision. The engine decides, per query, whether to search the web at all.
3. Answer variability. The same prompt can return a different answer on the next run.
4. Your own site and network. A deploy, a robots.txt edit, a CDN default or an index change.

The first three are outside your control. The fourth is where you can act. So the practical rule: run a check before and after every change you make, and after any vendor change you hear about. Our playbook on check cadence turns this into a schedule.

"ChatGPT" is not one knowledge state. OpenAI's model docs list a Feb 16, 2026 knowledge cutoff for the GPT-5.6 family and an Aug 31, 2025 cutoff for the Instant path (chat-latest), and they warn the snapshot "will be regularly updated." When routing sends a question to a different model, the answer can change by nearly a year of knowledge.

Retrieval is also discretionary. As of October 2026, ChatGPT search decides when to search, and OpenAI says it "rewrites your query into one or more targeted queries." Google says AI Overviews often do not trigger, and that AI Overviews and AI Mode "may use different models and techniques." Claude.ai search is a user toggle. See how often ChatGPT searches the web for the published estimates, which disagree widely.

What it means: a drop between two checks may reflect a retrieval-off answer, not a change in how the engine sees you. Before reacting, look at whether the answer showed citations.

These are the changes worth a re-check, because each can silently remove you from answers:

• Rendering changes. Vercel found ChatGPT and Claude crawlers fetch JavaScript but do not execute it. A redesign that moves pricing into client-side code can hide it from them.
• robots.txt edits. Rules are per bot. OpenAI states "each setting is independent," so blocking OAI-SearchBot removes ChatGPT search eligibility while GPTBot governs training only.
• CDN and firewall defaults. From 2026-09-15, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers. A site can lose access without anyone editing it.
• Snippet controls. A template that adds nosnippet or data-nosnippet keeps content out of AI answers.
• Index status. Google requires a page to be indexed and snippet-eligible to appear in AI Overviews or AI Mode.

Otterly's 2026 experiment shows how quietly index changes land. Google deindexed its two AI-generated test sites after about 14 and 90 days, with zero manual actions in Search Console. Google's AI surfaces stopped citing them, while Microsoft Copilot, which grounds on Bing, kept citing them. Check Google Search Console and Bing Webmaster Tools separately.

To test crawler access directly, the AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

Some differences between checks are artifacts, not changes:

• Different prompts. A tracker only sees answers to the prompts it asks. Change the prompt set and you change the result.
• Logged-in state. OpenAI says memory can shape how ChatGPT rewrites a search query. Check logged out with memory off.
• Method changes. Ahrefs reported 37.9% of AI Overview citations came from top-10 results in March 2026, but it changed its parsing method from its earlier study, so the two figures are not a trend. Any score, including a vendor's, can change method too.
• Blended scores. In Otterly's test, one engine supplied 69% of a site's tracked citations while four barely cited it. A single average would have hidden that.

Record the prompts, engines, date and method with every run. Setting a baseline before changes makes later checks readable.

1. Pick a fixed prompt set and a fixed list of engines, and keep both stable.
2. Run a baseline now. The free checker on this site gives a snapshot for up to three buyer questions.
3. Re-check after every deploy that touches rendering, robots.txt, CDN settings or meta robots tags.
4. Re-check when a vendor announces a model or default change.
5. Keep a regular interval for reporting, and say plainly that it is a reporting choice.

Do not re-check because you updated dates on old pages. Freshness as a citation lever is unresolved, and the main supporting study had no control group. Publishing more pages is not a shortcut either; see whether publishing more helps.

Is weekly or monthly better for AI visibility checks? No source answers this. Pick the interval your reporting needs, then add event-driven checks after site deploys, robots.txt or CDN changes, and vendor model announcements. A weekly check that misses a firewall change tells you less than one check run the day after that change shipped.

Why did my result change when I changed nothing? The engine may have routed your question to a different model, skipped web search for that run, or produced a different answer from the same prompt. A third party, such as a CDN applying new defaults, may also have changed your crawler access. Check whether the answer showed citations, then test crawler reachability before assuming anything about your content.

Do AI crawlers revisit my site on a schedule? No AI vendor publishes a crawl schedule. You can see actual visits in your server or CDN logs filtered to verified bots. Our page on [AI crawler frequency](/questions/ai-crawlers-crawl-frequency) covers what can and cannot be known about how often they come back.

Should I update content more often to stay visible? The evidence does not support updating content on a schedule to earn citations. Studies of freshness looked only at pages already cited and had no comparison group, and the dates involved are self-declared by publishers. Update a page when a fact on it changes, such as a price or product name, so engines that fetch it get the correct version.