Brand owners ask this before they invest in GEO, and it is the right question: if a new version of ChatGPT, Gemini or Claude ships next quarter, does the work have to start again? The honest answer is partly.

Some of what an assistant knows about your brand is fixed when the model is trained; some of it is fetched live when a shopper asks. The live part carries over between versions because it lives on the web and in your product data. The trained part refreshes on the model maker's schedule, not yours.

This article separates the two, explains what carries over, what can reset, and how an established brand keeps its visibility from quietly decaying between checks.

Where does an assistant's knowledge of your brand come from?

From two layers. The first is what the model learned during training, which stops at a cutoff date. The second is what it retrieves at answer time: web pages found through search, and product data from catalogs and feeds.

Shopping answers depend heavily on the second layer, because prices, stock and product ranges change faster than models are retrained. That is why current pages and data matter more than what a model remembers.

Model makers publish these cutoffs. Anthropic, for example, lists a "reliable knowledge cutoff" and a "training data cutoff" for every Claude model, and several model generations stay available at the same time, so different products and apps can be answering with different versions. Retrieval is what fills the gap: Google's AI Overviews and AI Mode draw on Google's index, and Shopify sends eligible products to AI channels through Shopify Catalog.

What carries over between model versions?

Everything that lives outside the model carries over: your crawlable pages, your product data, your reviews, and what retailers, press and customers have published about you. A new model reads the same web and the same catalog as the old one, so the work that made those sources clear and consistent keeps paying off.

• Your pages. Product facts, size guides, care and returns pages that answer buying questions.
• Your product data. What Shopify Catalog and your Google & YouTube channel send on your behalf.
• Outside descriptions. Reviews, stockist listings and press that describe you consistently.
• Working URLs. Addresses that still resolve when an assistant, or its sources, point to them.

What can reset or shift?

Three things can move without you doing anything: what a model remembers, how a new model weighs its sources, and the natural variation between one answer and the next. A fourth is self-inflicted: changes to your own store that break the links and facts assistants rely on.

• Memory. A model trained before your latest product line launched will not know it unless retrieval supplies it.
• Weighting. A new model or search system can prefer different sources for the same question.
• Variation. Answers differ run to run. A January 2026 SparkToro and Gumshoe.ai study found under a 1 in 100 chance that ChatGPT or Google's AI returns the same list of brands twice.
• Broken links. An Ahrefs study of 16 million URLs found AI assistants send visitors to 404 pages at 2.87 times the rate of Google Search. Renaming product handles without redirects makes this worse.

How do you keep visibility from decaying?

Treat the sources an assistant reads as living assets: keep facts current, keep addresses stable, refresh the pages that answer buying questions when your policies or products change, and re-check the same questions on a schedule. None of this depends on which model version is current.

1. Keep product facts current. Update materials, sizes and prices where they change, in the store and in your channel data at the same time.
2. Redirect, never strand. When you rename or retire a product or page, add a 301 redirect to the closest live equivalent. Say a brand renames its "Trail 2" boot to "Ridge" and the product handle changes: reviews, retailer pages and older answers still point to the old address, and without a redirect every one of those links ends on a 404.
3. Refresh the doubt-settling pages. A returns page that still shows last year's window can be quoted as if it were current.
4. Retire old names. Discontinued product lines left live on retailer pages can keep appearing in answers.
5. Re-check monthly. Ask the same buying questions several times each and compare with last month. This is the method.

Why is this ongoing work rather than a one-off project?

Because both sides keep moving. Your products, prices, policies and markets change every season, and assistants change their models, sources and shopping features several times a year. A one-off clean-up fixes today's gaps; a regular check notices the next ones before your customers do.

That is why the work we do for clients runs as an engagement rather than a single delivery: an agreed set of buying questions, a dated baseline, improvements the merchant approves before they go live, and a monthly check of what changed. Nobody can promise that a brand stays in AI answers. What can be done is to keep the evidence those answers rely on accurate and easy to find.

FAQ

Will a new AI model forget my brand?

It will not forget what it can retrieve. A new model reads the same web pages and product data as the previous one, so a brand with clear, consistent, crawlable sources carries over. What a new model can change is how it weighs those sources, which is why a check after a major model release is worth doing.

How quickly do assistants pick up changes to my store?

Changes reach retrieval once your pages are recrawled or your catalog data syncs, on the crawler's and the catalog's own schedules. They reach a model's built-in memory only when it is retrained, on the model maker's schedule. There is no dependable timetable for either, so verify with a repeat check.

Should I rewrite my content for every new model?

No. Model releases are not a reason to rewrite pages. The same principles hold across versions: accurate facts, pages that answer buying questions, consistent outside descriptions and working URLs.

How often should an established brand re-check its AI visibility?

Monthly for the questions that matter commercially, and again after a major change on your side, such as a new product line, new prices or a new market, or a major model release on theirs.

Sources

• Anthropic, Models overview (knowledge and training data cutoffs)
• Google Search Central, AI features and your website
• Shopify Help Center, Shopify Catalog and product discovery for agentic storefronts
• SparkToro, AIs are highly inconsistent when recommending brands or products (January 2026)
• Ahrefs, How often do AI assistants hallucinate links (September 2025)