Your category pages have an expiry date. Here is what the data shows.

Most retailers treat category and product pages as infrastructure. You build them once, update them for promotions, and leave the rest alone. That assumption is costing you visibility in every AI search engine operating today.

A reverse engineering study published by Resoneo in February 2026, based on several months of network traffic analysis and code decompilation of ChatGPT Search, uncovered something that changes how retail content strategy should be planned. ChatGPT does not search the web the way most people assume. It already has your older content baked into its training data. When it searches, it is specifically filling the gap with information it does not yet have, which means information published recently.

The practical consequence for retail is direct: a category page for "women's coats" last updated 14 months ago is not competing on equal terms with one refreshed last month. The AI is not looking for the most authoritative page. It is looking for the most current one.

This article explains the mechanism, what it means for product and category page strategy, and how to build a refresh cadence that keeps your catalog visible as AI search continues to grow.

The technical reason freshness matters more than you think

The Resoneo research reverse-engineered a component of ChatGPT Search called the Sonic Classifier, a probabilistic system that runs before any response is generated. Its job is to determine whether a query needs fresh external data or whether the model can answer from its trained knowledge alone.

The classifier assigns a search probability score. If that score exceeds a defined threshold (identified as 65% in the reverse-engineered code), ChatGPT triggers a live web search. If it stays below that threshold, the model responds from memory. For most product and category queries, the threshold is crossed. Which means ChatGPT is actively searching the web every time a user asks about a product category, a seasonal item, or a specific type of product.

But here is where it gets specific. The research also uncovered the recency filter system applied per query:

The model applies these filters to the web search results it retrieves. A query like "best winter coats for women 2026" will almost certainly trigger a 30-day recency filter. If your category page for winter coats was last updated in July 2025, it does not appear in those results. Not because it ranks poorly. Because it is filtered out before ranking even begins.

ChatGPT is not rebuilding its knowledge base when it searches. It is filling the gap between its training cutoff and today. Your content has to live in that gap.

This is a fundamentally different framing from traditional SEO, where an evergreen page accrues authority over time and continues ranking for years with minimal maintenance. In AI search, the freshness filter can override authority entirely. The question is not whether your page is good. It is whether it is current.

For a deeper look at how this shift from traditional SEO to AI search affects retail strategy overall, our piece on the end of search as we know it covers the structural changes in detail.

What counts as fresh content for an AI search engine

This is where most brands make a costly mistake. "Fresh content" in the context of AI search is not a new publication date on a page that hasn't changed. It is not a minor price update. It is not swapping a hero image. AI systems are reading and understanding content, not parsing metadata.

Fresh content, in practice, means content that has been substantively enriched since the model's last encounter with it. For a category page, that means:

• Updated editorial framing. A summer dresses category page should read differently in April than it does in February. Occasion references, styling notes, seasonal context. These signals tell the AI that the page is current and relevant to now.

• New product introductions with enriched descriptions. Adding new SKUs to a category with thin, promotional descriptions does not move the freshness needle. Adding them with full attribute sets, material details, and occasion guidance does.

• Refreshed SEO and GEO signals. Updated keyword emphasis reflecting current search trends, updated FAQ sections addressing current buyer questions, updated structured data reflecting current availability and pricing.

• Seasonal relevance signals. Content that explicitly references the current season, current trends, or current context performs significantly better in AI retrieval because it is legible as timely, not archived.

This connects directly to a broader point we have covered in our research on semantic coherence and AI search discoverability: AI systems reward pages where the content is consistent, complete, and coherent across every signal. A page refreshed for freshness but left incomplete on attribute coverage gets filtered for freshness but still fails on comprehensiveness.

Why seasonality is now an AI search signal

Retail has always operated on a seasonal calendar. What has changed is that the seasonal calendar now maps directly onto AI search behavior, not just conversion rates.

The fan-out system identified in the Resoneo research reveals that a single user query triggers multiple parallel searches across different indexes. A query for "linen trousers for summer" might generate simultaneous searches across web content, product indexes, and news. Each of those parallel searches applies its own recency filter.

For a retail brand, this means the window for seasonal content to be AI-visible is narrower than the window for that content to be commercially relevant. A category page optimised for summer should be refreshed and indexed before the AI recency filter reaches the point where it deprioritises it. In practice, that means refreshing six to eight weeks before the season is commercially active, not at the point of launch.

Consider a typical seasonal arc:

The brands currently winning at AI search have, in many cases, not made a single large investment. They have made a consistent one. Quarterly refreshes across high-value category pages, applied systematically, compound over time in a way that single-burst publishing strategies cannot replicate.

Product pages are not exempt

The instinct in most ecommerce teams is to focus freshness efforts on category pages and editorial content, treating product pages as a fixed asset. That instinct is increasingly wrong.

For AI assistants making product recommendations, the product page is the primary citation source. When ChatGPT recommends a specific product, it draws on the most complete, most current description it can find. As we documented in our research on how enterprise catalogs lose AI visibility through attribute gaps, most enterprise catalogs publish 5 to 7 attributes per SKU when AI recommendation logic needs 15 to 20. Refreshing those descriptions quarterly, adding new attribute layers, and updating them to reflect current availability and seasonal context keeps them inside the recency window that AI search prefers.

There is also a compounding effect. Product pages that receive regular content updates generate fresh crawl signals. They are more likely to be re-indexed recently. They are more likely to appear within the 30-day recency filter for product queries. This is not a theoretical SEO gain. It is a direct pathway to AI citation frequency.

Brands moving from AI writing tools to content infrastructure, as we discussed in our piece on why retail content teams are shifting to AI infrastructure, are precisely the ones building the operational capacity to refresh at this cadence. A team using a generic AI tool can produce a description once. A team with content infrastructure built on brand-trained models can refresh 10,000 SKUs quarterly without a proportional increase in resource.

The multilingual freshness gap

One dimension of content freshness that almost no brand has addressed systematically is the multilingual one. The recency filter applies per market and per language. ChatGPT in France is searching for French-language content. If your French category pages were last refreshed 18 months ago while your English ones were updated last month, your French catalog is outside the recency window even if your English catalog is not.

This creates an invisible discoverability gap. You cannot see it in your analytics because AI-referred traffic has no reliable attribution path in most platforms. But it is real, and it widens every quarter that multilingual pages go unrefreshed.

As we explored in the brand is no longer in the room, customers in markets you cannot monitor are asking AI assistants about your products in languages you do not speak. If the content available in those languages is stale, AI will pass your brand over in favour of a competitor whose local content is current.

Building a quarterly refresh system

A content refresh strategy is not a one-time campaign. It is an operational rhythm. Here is a framework for building it:

1. Tier your catalog by commercial value and AI query volume. Not all pages need quarterly refreshes. Your top 200 category and subcategory pages, and the product pages that drive 80% of revenue, should be refreshed every quarter. Long-tail pages can operate on a twice-yearly cycle.

2. Build a seasonal refresh calendar and start early. Map your category refresh schedule to the seasonal calendar but push each refresh six to eight weeks earlier than your commercial launch window. The AI needs time to index and weight your content before the peak query window opens.

3. Define what a refresh actually means. A refresh is not a cosmetic edit. For a category page: update the editorial intro, refresh seasonal references, add new arrivals with full attribute descriptions, update the FAQ section. For a product page: expand the attribute set, add occasion and styling notes, update availability signals and structured data.

4. Refresh multilingual pages on the same schedule. Not a month later. The same schedule. If your English category pages refresh in late January for spring, your French, German, Italian, and Spanish equivalents refresh in the same window. Multilingual content freshness is not a separate workstream. It is the same workstream, executed across more languages.

5. Instrument the output. Track crawl frequency per page in Google Search Console before and after a refresh cycle. Monitor AI citation frequency by asking the main AI assistants about your product categories monthly. The uplift from systematic refreshing is measurable within 60 to 90 days.

For a complete view of how content structure, attribute depth, and semantic clarity interact with AI search performance, the 2026 State of AI Search report covers the full picture with current benchmark data.

Freshness is a structural advantage, not a maintenance task

The brands that understand the Sonic Classifier's recency bias and build their content operations around it are not doing more work than brands that don't. They are doing the same work on a different schedule, with a different operational infrastructure underneath it.

The competitive advantage compounds quietly. A category page refreshed every quarter accrues freshness signals, crawl frequency, and AI citation history in a way that a static page cannot replicate regardless of its original quality. The AI search landscape rewards consistency over perfection.

The question is not whether your content is good enough. It is whether your content is current enough, and whether you have the infrastructure to keep it that way across every category, every product tier, and every language you operate in.