Free Knowledge Cutoff Check: Is Your Brand News After the AI Cutoff?

99Visibility Research · Updated October 21, 2026 · 5 min read

Short answer

A knowledge cutoff check compares the dates of your launches, price changes and rebrands with each model's published training cutoff. OpenAI lists Feb 16, 2026 for the GPT-5.6 models and Aug 31, 2025 for ChatGPT's Instant path. Any fact dated after a model's cutoff can only reach that model's answers through live search. So the fix for newer facts is access: crawlers must be allowed, reachable and able to read the page.

Key takeaways

List your dated brand facts, then mark each one as before or after every model's published cutoff.
As of October 2026, OpenAI lists a Feb 16, 2026 cutoff for GPT-5.6 Luna, Sol and Terra, and Aug 31, 2025 for the Instant path (chat-latest).
Facts after a cutoff reach answers only through search, so robots.txt, firewall access and server-rendered HTML decide whether they can be found.
A fact before the cutoff is not guaranteed to be known, and nobody can measure what made it into training.
Search fires at the model's discretion, so a reachable page can still lose to an older trained answer on any given question.

What the free check tells you

The check sorts your brand facts into two groups. Facts dated before a model's cutoff could be in its trained memory. Facts dated after it cannot be, so the model only sees them when it searches the web.

That split matters because the two groups need different fixes. An old fact stuck in training is a waiting game. A new fact the model never sees is usually an access problem you can fix this week.

If you want the background first, read why AI knows your old news. This page is the hands-on version.

How to run the check by hand

List your dated facts. Write down every change a buyer would care about, with the date it went public: product launches, price changes, new plans, a rebrand, an acquisition, a new HQ, a discontinued feature.
Look up each model's cutoff. Vendors publish these on their model documentation pages. As of October 2026, OpenAI's model pages list a Feb 16, 2026 knowledge cutoff for gpt-5.6-luna, gpt-5.6-sol and gpt-5.6-terra. The non-reasoning Instant path, chat-latest, lists Aug 31, 2025. Note the date you checked, because OpenAI warns these snapshots are regularly updated.
Mark each fact. Put it in a simple table: fact, public date, and before or after each cutoff. Say you run a dental clinic in Austin that raised prices in March 2026. That change is after both OpenAI cutoffs above.
Ask the question without search. Use a logged-out or temporary chat so memory does not shape the answer. If the reply has no citations, treat it as trained knowledge. Compare it with your table.
Check robots.txt for search bots. Open https://yourdomain.com/robots.txt and confirm the retrieval bots are not blocked. A minimal allow rule looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

Check the fact is in raw HTML. Run curl -s https://yourdomain.com/pricing | grep "your new price". If the grep finds nothing but the browser shows the price, JavaScript is loading it.
Check your firewall or CDN. Repeat the curl with a crawler user-agent (the exact strings are on each vendor's crawler page) and look for a 403, 429 or a challenge page.

How to read the result

Each fact lands in one of four states.

Before the cutoff, answer correct. Nothing to do for that fact.
Before the cutoff, answer wrong or old. The model may have learned an earlier version. This is slow to change, and how long AI takes to update brand info covers the timing.
After the cutoff, answer old. Expected. The model is answering from memory. Your job is to make the new fact findable when it does search.
After the cutoff, missing even with search on. This points at access. Work through steps 5 to 7 again.

One trap: "ChatGPT" is not one knowledge state. The Instant path's cutoff is about five and a half months older than the GPT-5.6 models. Two people asking the same question can get answers built on different cutoffs. For a newer brand, whether ChatGPT can know a new company explains what that means in practice.

Why access is the fix for post-cutoff facts

Nothing you publish this quarter reaches an answer given without search. Search is the only route for new facts, and it has three gates you control.

Per-bot permission. OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. GPTBot is a separate setting that governs training only.
Network reachability. OpenAI's help centre says to allow OAI-SearchBot and confirm your host or CDN allows traffic from its published IP addresses. A firewall can block what robots.txt allows.
Rendering. Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. Put prices, plan names and dates in the server-rendered HTML.

Fixing these gives you eligibility, not a citation. The step-by-step repair is in correcting outdated facts past a knowledge cutoff.

What this check cannot tell you

Whether a pre-cutoff fact was learned. Nobody, including the vendors, publishes training membership. Being before the cutoff is a ceiling, not a guarantee.
Whether a given answer used search. ChatGPT searches at the model's discretion, and Claude.ai search is a user toggle. Without visible citations you don't know which path produced the answer.
Which model a plan runs. Plan-to-model mappings come from secondary reports, not OpenAI. Treat any mapping as provisional.
Other platforms. Each vendor has its own cutoff and search behaviour. Run the same table for what Claude says about your brand and the rest.

The trained knowledge vs search split for ChatGPT goes deeper on which errors you can fix and which you wait out.

The faster way to see the result

The manual table tells you which facts are at risk. It doesn't show what the engines say today. The free checker on this site runs up to three buyer questions and scores the answers, so you can see whether your newer facts appear. It is a snapshot, not ongoing tracking. If you want repeat audits later, you can start on the Free plan with no card.

Frequently asked questions

Does allowing GPTBot get my new facts into the next model?

Nobody can say. OpenAI describes GPTBot as the crawler for training data and claims no search benefit from allowing it. Whether training inclusion changes what a future model knows about your brand is unmeasured, and any effect would only show up a full model generation later. For facts you need in answers now, OAI-SearchBot permission and reachability are the controls that matter.

Does a knowledge cutoff affect Google AI Overviews?

AI Overviews draw on Google's search index rather than only trained memory. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link, with no extra technical requirements. Googlebot renders JavaScript, so the rendering gap matters less there. Check indexing in Search Console, and remember there is no robots.txt token for AI Overviews alone.

How often should I rerun the cutoff check?

Rerun it whenever a vendor ships a new model or updates a cutoff, and whenever you change a buyer-facing fact such as pricing. OpenAI notes its model snapshots are regularly updated, so a date you recorded in one month may be out of date the next. Keep the date you checked next to each cutoff in your table.

If I block AI crawlers, will old facts disappear from answers?

No. Blocking search bots removes your pages as a source for live answers, but it does nothing to what a model already learned in training. The old version can keep appearing from memory, now with no fresh page to correct it. The tradeoffs are covered in [whether AI can still talk about your brand if you block crawlers](/questions/block-ai-crawlers-brand-still-mentioned).

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.