Every few months, Google announces an upgrade to Gemini — their underlying AI model — and the coverage that follows tends to fall into one of two camps. Either it’s breathless enthusiasm about the dawn of a new advertising era, or it’s dismissive cynicism from people who’ve heard too many announcements that didn’t change anything.

The truth, as usual, is more practical than either.

Here’s what actually changes for your e-commerce business when Google improves its AI — and, just as importantly, what doesn’t.

What Gemini Actually Powers in Your Campaigns

Before getting into what a smarter model means, it’s worth being clear about what Gemini is doing in your account right now.

Gemini powers the decision-making layer across most of Google’s automated ad products. In Shopping campaigns, it determines which of your products to show for a given search query and how aggressively to bid for that placement. In Performance Max, it decides which combination of assets to serve, which audiences to prioritise, and where across Google’s network to show your ads. In Search campaigns using broad match, it decides which queries are “close enough” to your keywords to trigger your ads.

Every time Google improves Gemini, these decisions get sharper. The model becomes better at predicting which product a searcher is likely to buy, which user profile tends to convert, and which queries represent genuine purchase intent versus idle browsing.

What Actually Changes With a Better AI Model

The most meaningful improvement, in practical terms, is relevance matching in Shopping.

Google’s Shopping algorithm needs to understand what your products are in order to serve them against the right searches. A smarter AI is better at parsing your product titles and descriptions, understanding their category and attributes, and connecting them to the specific queries where a purchase is likely.

This means a well-structured product feed gets rewarded more as the AI improves. A product titled “Leather Wallet — Brown — Slim — RFID Blocking — Men’s” gives the algorithm considerably more to work with than one titled “Wallet 001.” With an older, less capable model, both might surface for “men’s leather wallet.” With a more capable one, the well-described product tends to win more of the relevant placements — and fewer of the irrelevant ones.

Better AI also improves Performance Max in a meaningful way. The system gets more accurate at predicting which creative assets will resonate with which audience segments, which means your campaigns can drive better ROAS from the same budget without you changing a single campaign setting.

What Doesn’t Change

Here’s the part that gets glossed over in most coverage: a better AI model cannot fix bad inputs.

Garbage in, garbage out — it’s an old principle but it’s never been more relevant than in AI-powered advertising.

If your product feed has wrong categories, vague titles, missing attributes, or out-of-date pricing, the AI has less to work with. A smarter model hitting a poor feed will still underperform against a less sophisticated model hitting a well-structured feed. The AI raises the ceiling; it doesn’t do the groundwork for you.

Conversion tracking is the other non-negotiable. Smart Bidding — and every Google AI system that optimises towards conversions — learns from the conversion signals you send it. If your tracking fires on the wrong page, fires twice, or doesn’t fire at all, the AI is learning from corrupted data. It will optimise towards the wrong behaviour, and a smarter AI will do this more efficiently and confidently than a less capable one. That’s not better.

The Inputs That Determine Your Ceiling

What this means practically is that your job, as the person managing Google Ads for an e-commerce business, is to give the AI the best possible inputs — because that’s where you have control.

Product titles that reflect how people actually search. If someone searches “navy blue linen trousers women” and your product is titled “Ladies Trouser — Style 42B — Navy,” the algorithm has to work much harder to make the connection. Rewrite titles to include the attributes customers actually use in searches.

All required and recommended attributes populated. Google’s product specification includes dozens of optional attributes — colour, size, material, pattern, gender, age group. Each one you fill in gives the AI more signal. The accounts I see underperforming in Shopping are almost always missing attributes that competitors have populated.

Conversion tracking firing correctly on the order confirmation page. Not the basket page. Not the checkout page. The order confirmation page, once, per transaction. Verify this with a test purchase before assuming it’s correct.

The Practical Action

Before the next time Google announces a Gemini upgrade, run a quick feed audit. Pick ten of your worst-performing products — the ones with impressions but low conversion rates — and look at their titles and descriptions with fresh eyes. Would someone searching for that product recognise it from the title alone? If not, that’s your starting point.

A better AI model genuinely is better. But it rewards the businesses that have done the basics well. Those that haven’t will find that improvements to Google’s AI change very little for them.

If your product feed isn’t in the shape it needs to be, take a look at how Roksys approaches product feed management — it’s where most of the untapped performance in Shopping campaigns is hiding.