NIQ and Similarweb have announced a joint system for measuring purchases, with artificial intelligence doing a significant part of the journey. It is no longer just about the user’s request and the link in the answer: the AI assistant can find the product, compare the options, form a recommendation and guide the person to the order. For the advertising market, this creates a new problem of attribution – the brand sees the sale, but does not always understand where exactly the demand arose and why the algorithm showed one product instead of another.
What NIQ and Similarweb announced
The joint solution will combine NIQ’s product and sales analytics with Similarweb’s digital behavior data. Companies plan to track five consecutive levels: visibility of the product in AI-answers, correctness of presentation, transition to the site, conversion and actual sale. The initial version is expected to arrive in the fourth quarter of 2026 for a limited set of yet-to-be-named categories and markets.
- Q4 2026— the announced date of launch of the first version.
- Five groups of signalsmust connect AI product detection with a commercial outcome.
- ChatGPT, Gemini, AI Mode, Perplexity and Claudementioned in the profile review as environments intended to be covered.
- Categories, markets and priceof the product at the time of announcement are not disclosed.
Why ordinary web analytics is no longer enough
The classic digital funnel begins with an impression or search request, continues with a click and ends with an action on the site. In agent commerce, part of this sequence takes place inside another’s interface. A person may not visit a brand page until the assistant has sifted through most of the alternatives. In such a model, the position of the product in the answer becomes similar to an organic issue, a seller’s recommendation and an advertising shelf at the same time.
This means that a brand should measure more than just traffic. It is important to know whether the product falls into the set of candidates, what characteristics the model cites, whether it does not confuse packaging and options, how often it recommends a competitor and what sources it uses for justification. If product data is incomplete or inconsistent, additional advertising budget may not correct the problem at the algorithmic selection stage.
What’s changing for media planning
Advertisers will have to add another layer to the usual channels — AI-discovery. This is not necessarily a separate media budget already today, but definitely a separate area of responsibility. SEO, e-commerce, content, retail media and brand marketing teams must work with a single product description and agree on which signals to consider AI influence.
The issue of incrementality will be especially difficult. If a person saw a TV ad, then asked an assistant and bought a product from a marketplace, that last digital contact doesn’t tell the whole story. The new system will only be useful if it doesn’t announce AI as the sole reason for a sale, but shows its contribution alongside other contacts.
Where the risks remain
The announcement describes a direction, not a ready-made universal standard. The companies did not disclose the methodology for matching AI responses to sales, the frequency of data collection, the rules for working with personalized results and the way closed purchases are taken into account within the assistant. It is also unknown how fully the platform will allow external meters to see the user’s path.
There is also a conflict of incentives. Measurement providers are interested in proving that the new channel is large and manageable, and the advertiser needs verification that the metrics are truly predictive of incremental sales. Therefore, the first dashboards should be interpreted as a diagnostic tool, and not as the final currency of the market.
What the brand should prepare now
- to unify the names, characteristics, prices and availability of goods in own and partner sources;
- check how the main AI services describe the brand and which pages are linked;
- share visibility, referral traffic, conversion and confirmed sales;
- establish control groups and experiments so as not to confuse correlation with growth;
- save the history of answers, because model recommendations change without the usual ranking of positions.
We have already sortedhow technological mechanics become part of the brand experience, and explainedwhy a media metric cannot be automatically equated to a business effect. The measurement of agent commerce combines these two problems: the algorithm both shapes the experience and makes it difficult to prove the result.
Conclusion UA Banner
The partnership between NIQ and Similarweb shows that AI assistants are moving from the category of experimental interface to the status of a separate commercial environment. What will be most valuable is not a beautiful map of brand mentions, but the relationship between the quality of product data, model recommendation and actual sales. Until a transparent methodology appears, advertisers should test the channel, but not allow the new metric to prove its own value.
Sources:official announcementNIQand an independent profile reviewPPC Land. The description of the features refers to the claimed product, the first version of which has not yet been available for independent testing.
