Quarterly AI Visibility Review Checklist: Five Checks to Repeat Every Three Months

99Visibility Research · Updated October 21, 2026 · 8 min read

Short answer

A quarterly AI visibility review is a five-part check you run every three months. Re-run the access audit, review which model knowledge cutoffs changed, refresh the competitor set, re-validate your tracked prompts against real grounding queries, and drop tasks the evidence does not support. Access comes first because it breaks silently. The review cannot guarantee citations. It removes blocks and wasted work.

Key takeaways

Run the access audit first: robots.txt per bot, firewall reachability, rendered versus raw HTML, snippet controls and index status in both Google and Bing.
Model cutoffs differ by model and route, so list what changed about your brand since each cutoff and expect stale answers when retrieval is off.
Engines rewrite prompts into their own queries, so check your tracked prompt set against Bing Webmaster Tools grounding queries every quarter.
Cut tasks with no evidence behind them, such as adding schema or llms.txt to get cited, and spend that time on access and accuracy.
Compare scores across quarters only when the prompt set, engines, retrieval state and scoring formula stayed the same.

Why AI visibility needs a quarterly review

AI visibility can break without anyone touching your site. In the last quarter alone, Cloudflare changed its default AI crawler settings on 15 September 2026, and Perplexity retired its Sonar Chat Completions API on 27 September 2026. Either change can affect whether an engine reaches you or how a tracker measures you.

A quarter is the right interval for the full review. Monthly is too often to read a pattern from a prompt sample. Once a year is too slow to catch a firewall that started turning away a search crawler.

The review has five parts, in a fixed order. It is built around what you control. A critical survey of 45 GEO studies (arXiv:2607.14035, July 2026) found that no reviewed technique shows a stable, cross-platform causal effect on discoverability. The levers that hold up are access levers, plus getting your facts right. If this is your first pass rather than a repeat, run a full AI SEO audit once before you start the quarterly cycle.

Step 1: Re-run the access audit

Access is the gate every engine passes through before it can quote you. Re-check it every quarter, in this order.

Resolve robots.txt bot by bot. OpenAI documents that OAI-SearchBot controls ChatGPT search and GPTBot controls training only, and that each setting is independent. Check OAI-SearchBot, PerplexityBot, Claude-SearchBot, Meta-WebIndexer, Googlebot and bingbot. Look for a blanket AI block someone added since last quarter.
Request a key page as each crawler. A firewall or CDN can return 403 to a bot that robots.txt allows. OpenAI's help center says eligibility needs both robots.txt permission and a host or CDN that allows its published searchbot IP addresses. On Cloudflare, review your AI crawler categories against the 15 September defaults. The AI Crawler Checker requests a page as each AI crawler and shows which ones the firewall turns away.
Diff raw HTML against the rendered page. Vercel found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. New pricing tables and feature pages often ship client-side only. Googlebot renders, so this check matters for the other engines.
Scan templates for suppression. Look for noindex, nosnippet, max-snippet:0 and data-nosnippet. Google and Bing document these as controls that keep content out of AI answers. A tag meant for a cookie banner can hide a pricing block.
Check index status engine by engine. Google says a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode. In Otterly's 2026 experiment, Google deindexed two AI-generated sites while Bing kept them, and Microsoft Copilot kept citing them. A healthy Google index tells you nothing about Bing, and the reverse.
Check the generative AI control in Search Console if your property shows it. Inclusion is the default, but child properties inherit from the parent, so an opt-out can arrive silently.

The ChatGPT visibility checklist covers the OpenAI-specific parts of this step in more detail.

Step 2: Review model cutoff changes

When an engine answers without searching, it answers from training data frozen at a cutoff. Nothing you publish this quarter reaches those answers. That changes only with the next model generation.

Cutoffs also differ inside one product. OpenAI's model docs (verified August 2026) list a 16 February 2026 cutoff for the GPT-5.6 family and 31 August 2025 for the chat-latest Instant path. "What ChatGPT knows" is not one value.

List the current models for each engine you track, with each published cutoff.
List what changed about your brand since each cutoff: prices, product names, launches, ownership, locations.
For each change, ask the engine about it. Note whether the answer cited sources (retrieval on) or not (retrieval off).
Where a retrieval-on answer is still wrong, check that the correct fact is stated plainly, once, in HTML a crawler can fetch. A specific fact on a fetchable page gives the engine a sentence it can quote.
Fix contradictions on your own site first. If the pricing page and the footer disagree, an engine has two answers to choose from.

Step 3: Refresh the competitor set

Your competitor list goes stale faster than your product does. The brands engines name for your buyer questions this quarter may not be the ones you picked a year ago.

Re-run your core buyer questions on each engine and record every brand named.
Drop competitors that no longer appear. Add the ones that now do.
Record which third-party sites the engines cite for those questions. When engines cite other sites instead of yours, being named on those sites is associated with appearing in the answer. That is an association, not a guaranteed effect.
Compare yourself with brands of similar size. Brand prominence is the biggest uncontrolled factor in citation studies, so a benchmark against household names mostly measures brand size.

Say you run a hypothetical regional accounting firm. Your useful comparison set is other regional firms engines name, not the national chains that appear in every answer.

Step 4: Re-validate the prompt set against grounding queries

Engines don't search the prompt you type. OpenAI says ChatGPT search rewrites a query into one or more targeted queries. Google may fan a question out across subtopics. The Gemini API generates its own searches. Your tracked prompts are a proxy for an input you can't see.

Bing Webmaster Tools is the exception. Its AI Performance report shows grounding queries: the phrases the AI used when retrieving content it cited in Copilot and Bing answers. It had no API when last checked in August 2026, so export from the interface.

Export grounding queries for the quarter.
Map each tracked prompt to the grounding queries it resembles.
Add prompts for topics that appear in grounding queries but not in your set.
Retire prompts that never map to anything real.
Check how your tracker runs each engine. Record whether retrieval was on. A tool that called Perplexity Sonar without migrating may have returned ungrounded answers since 27 September.

The Microsoft Copilot visibility checklist explains how to read the rest of that report.

Step 5: Prune tasks the evidence does not support

Most AI visibility backlogs carry tasks inherited from old blog posts. Each quarter, cut the ones primary sources contradict.

Task on the list	What the evidence says
Add schema or FAQ schema to get cited	Ahrefs' matched test found AI Overview citations fell 4.6% after adding schema. FAQ rich results were deprecated in May 2026. Existing markup does no harm, so leave it, but stop treating it as a lever
Publish an llms.txt	Google says Search ignores it. No engine documents reading it
Improve the E-E-A-T score	Google says E-E-A-T isn't a specific ranking factor. There is no score
Chunk or rewrite content for AI	Google says you don't need to write in a special way and there's no ideal page length
Ping IndexNow so AI cites you faster	The spec never mentions AI. A 200 response confirms receipt only
Track your AI rank in Google	Google's AI report exposes impressions only. No position data exists

Also check that nobody added hidden text or instructions aimed at AI crawlers. Injected content is illegitimate, and engines and search operators treat manipulation as a risk, not a tactic. If a vendor's deliverables are mostly items from this table, the tracking tools versus fixing tools comparison explains what to ask for instead.

The quarterly checklist

Copy this list into your review doc each quarter.

robots.txt resolved for OAI-SearchBot, PerplexityBot, Claude-SearchBot, Meta-WebIndexer, Googlebot and bingbot
GPTBot decision recorded as a training choice, separate from search
Key pages requested as each crawler, status codes logged
CDN and firewall AI crawler settings reviewed
Raw HTML contains pricing, product names and key facts before JavaScript runs
No unintended noindex, nosnippet, max-snippet:0 or data-nosnippet
Index status checked in Google Search Console and Bing Webmaster Tools
Search Console generative AI control checked, including inherited settings
Current model cutoffs listed per engine
Brand changes since each cutoff listed and spot-checked
On-site fact contradictions fixed
Competitor set refreshed from real answers
Frequently cited third-party sites recorded
Grounding queries exported and mapped to the prompt set
Retrieval state recorded for every tracked engine
Unsupported tasks removed from the backlog

For the one-off version of this list, see the AI SEO checklist.

What good and bad results look like, and how often to repeat

Good results

Every search crawler you allow gets a 200 on your key pages.
Key facts are present in raw HTML.
Index status is healthy in both Google and Bing.
Most tracked prompts map to real grounding queries.
The backlog got shorter, not longer.

Bad results

robots.txt allows OAI-SearchBot but the firewall returns 403.
Pricing appears only after JavaScript runs.
A score moved and you can't say whether retrieval was on.
The competitor set hasn't changed in a year.

A score change between quarters means something only if the prompt set, engines, retrieval state and scoring formula stayed the same. Note every change to those in the review doc. The free checker on this site gives a snapshot for three buyer questions, which is useful as a quick comparison point, not as a trend.

Cadence

Run the full review every quarter. Re-run Step 1 alone after any CDN change, template change or major deploy. Run Step 2 out of cycle when an engine announces a new default model.

Frequently asked questions

Should my AI visibility score go up every quarter?

Not necessarily, and no one can promise it will. You control access, the facts on your site and how you measure. You don't control what models say or when retrieval fires. A flat score with a clean access audit and corrected facts is a better quarter than a rising score you can't explain. Judge the review by what it fixed.

What if I don't have Bing Webmaster Tools set up?

Set it up before next quarter. It is the only first-party source of the queries AI systems actually ran when citing pages, and it also shows Bing index status, which Copilot answers draw on. Until then, treat your prompt set as an unvalidated proxy and say so when you report results internally.

Does this replace a regular SEO audit?

No. Index eligibility overlaps, but an AI review adds per-bot access, rendering for crawlers that don't run JavaScript, retrieval state and factual accuracy. An SEO audit covers ranking and site health for classic search. The differences are laid out in [how an AI visibility audit differs from an SEO audit](/questions/ai-visibility-audit-vs-seo-audit).

Who should own the quarterly review?

One owner, usually in marketing, with an engineer on call. The access audit needs someone who can read robots.txt, CDN settings and raw HTML. The cutoff, competitor and prompt steps need someone who knows the product and the buyers. Split the work, but keep one review doc so changes between quarters stay visible.

Sources

This page is AI-assisted. It was written from a verified evidence base and checked against our editorial policy. Engine behaviour changes often; facts are as of October 21, 2026.