Written by: Nicole Ogonowska, IT Growth Manager, Digital Colliers
Google's AI Mode connected apps rollout changes the shape of product discovery in a way most mid-market retailers aren't set up for. When a shopper asks the assistant to find a navy waterproof in their size under £150, the model doesn't crawl your site. It reads whatever feed you've handed it, or whatever your connected app returns in real time. If that data is stale by twelve hours, you're invisible for twelve hours. Your competitor whose feed refreshes in minutes gets the impression, the click, and the sale.
This isn't a future problem. It's a now problem, and it lands on top of a market that grew roughly 3% in the UK last year. Single-digit growth is the new baseline, so every discovery surface you lose hurts more than it used to.
Why AI-mediated discovery punishes stale feeds
Traditional SEO tolerated latency. A product page could sit in the index for days between crawls and nobody really noticed. AI Mode doesn't work like that. The assistant needs to answer a question in one shot with items that are actually buyable right now. So the ranking signal shifts toward two things: is this data fresh, and is it structured well enough that the model can trust it.
What that means in practice:
- Price, stock, and availability need to reflect the last few minutes, not last night's batch
- Variant-level data (size, colour, fit) has to be complete, because the assistant filters on it before it even considers your listing
- Structured attributes (material, care, dimensions, compatibility) determine whether you match long-tail intent at all
If your PIM exports a nightly CSV to Merchant Center and calls it done, you're competing with one hand tied.
What needs to change in the PIM-to-feed pipeline
The operators I see adapting well are treating the feed as a live product, not a scheduled export. A few shifts show up again and again:
- Event-driven updates from the source of truth. When stock changes in the OMS or a price changes in the PIM, an event fires and the feed layer picks it up. No waiting for the 2 a.m. cron.
- A feed cache with per-attribute TTLs. Price and stock refresh in minutes. Descriptions and imagery can refresh daily. You don't need to rebuild the whole feed to push a stock update.
- Variant expansion at write time, not read time. Every size and colour combination gets its own row with its own availability, because that's what the assistant is going to filter on.
- Validation before publish. Missing GTINs, empty size fields, and broken image URLs get caught and quarantined instead of shipped.
None of this is exotic engineering. It's mostly about deciding that your feed pipeline is a first-class system and staffing it that way.
The measurable checks that tell you if you're competing
You can't manage feed quality by vibes. The teams shipping this well run a small dashboard against three questions.
- Feed lag: median and p95 time between a change in the PIM or OMS and that change appearing in the outbound feed. If p95 is measured in hours, you have work to do.
- Inventory accuracy: percentage of feed rows where the stated availability matches actual warehouse state at read time. Below 98% and you're paying for clicks on things you can't ship.
- Variant coverage: percentage of live SKUs with complete variant-level attributes (size, colour, GTIN, image). Gaps here are why the assistant skips you for a competitor who filled them in.
Layer on category-specific checks too. If you're in apparel, where UK return rates run 25 to 40% depending on category, size and fit accuracy in the feed directly affects returns downstream. A shopper who gets matched to the wrong variant is a return waiting to happen.
The left-behind risk is quiet
The uncomfortable part is that you won't get an alert when this starts costing you. There's no dashboard tile that says "AI Mode skipped you 4,000 times today." You'll just notice, over a quarter or two, that your paid channels are working harder for the same revenue. CAC across DTC has already risen roughly 40% since 2023, and Meta CPMs kept climbing through 2024 and 2025. Organic AI discovery is one of the few surfaces where the economics still favour the retailer, but only if the assistant can actually read your catalogue.
If roughly 30% of SKUs at a typical multi-channel brand already lose money per order after returns and ad spend, ceding fresh-feed discovery to competitors is not a cost you can afford to add on top. The fix isn't glamorous. It's a pipeline problem, and pipeline problems get solved by treating them like products.