GetRaight Analyse my site

Our research

What we measured

Before launching GetRaight, we read hundreds of online stores and asked AIs about their products, then checked every result by hand. Here is our evidence, with its sources.

Found on real sites

Real errors, with their cause and their fix

We asked three AIs about the products of five French online retailers, then reopened every page by hand to check. Here is what we found.

9 in 10Products on sale reported as “out of stock”On the site of a French accessories brand, all three AIs answer that 9 products out of 10 are not in stock. Every one of them can be bought. The cause: an “Out of stock” heading left in the code of each page, invisible on screen. The fix takes a few lines.
+8%A price inflated by AIOn the site of a French skincare brand, two AIs out of three quote €24.51 for a product listed at €22.61. A customer who compares thinks your product costs more than it does, and may buy elsewhere.
10 in 10Every confirmed error was accurateGetRaight only calls an error “confirmed” when at least two AIs out of three give it and the reference was read on screen, as a customer would see it. The 10 confirmed errors of this test were checked by hand: all were true.

Test run on 15 November 2026: 5 French online retailers, 10 products per site, 180 questions asked to three AI models (OpenAI, Google, Perplexity) through Perplexity. Brands are not named.

Findings on 14 accommodation and service sites

Security: what AIs may pass on to your customers

We read 20 pages on each of 14 French sites: 8 hotels and campsites, 6 home services and tradespeople. Every gap below was checked by hand.

3 of 4When the site says nothing, AIs answer anywayWe put 4 customer questions to 4 AIs on 10 of these sites. Where the site did not give the information, 33 answers out of 44 still give a precise one: check-in time, opening hours, phone number, cancellation. Taken elsewhere, or guessed.
11 of 40Basic information missing from the siteOn these 10 sites, 11 of the 40 pieces of information asked for appear nowhere in the pages read: cancellation terms, check-in time, opening hours, phone number.
2 of 3Free cancellation promised in errorA campsite refunds no cancellation without insurance. Yet two AIs out of three said you could cancel free of charge up to 30 days before arrival. Twenty minutes later, all three gave the right answer.
3 of 8No cancellation terms across 20 pagesNot on the home page, the rooms, rates or contact pages. A customer who asks an AI whether they can cancel will not get the property's answer.
6 of 8Check-in time missing from the machine-readable formatIt is sometimes written on the page, but not in the structured data, the part of the code made for machines. The AI has to find it in the text, or guess.
3 of 8An English home page for the French marketA crawler that does not state its language gets the English version. Answers given to a French customer start from that text.
6 of 6No opening hours in the machine-readable formatPlumber, cleaning, childcare, driving school: where hours are written, it is for visitors: none of the six declares them for machines.
4 of 6No phone number in the machine-readable formatWhere the number is shown, the AI has to spot it in the text. One misreading, and the customer calls a wrong number, or a competitor.
2 of 6No response time stated“How soon can you come?” is often the first question. Two sites out of six do not answer it anywhere.

Test run on 16 November 2026 on 14 French sites (8 hotels and campsites, 6 services), 20 pages read per site, sites not named. On 10 of them, 4 questions put to four AI models (OpenAI, Google, Anthropic, Perplexity) through Perplexity's API, i.e. 160 answers compared with the site. The campsite case comes from a first run, the same day, with three models.

Our measurement

We analysed 549 online stores. Here is what is missing most often.

55%of sites do not state anywhere on their product pages a delivery time an assistant can find.
58%do not say how many days a customer has to return a product.
89%do have the technical basics in place. What is missing is the answers to shoppers' questions.

GetRaight measurement, November 2026: 430 sites in Europe and 119 in the United States, up to three product pages per site.

What you stand to lose

A customer misinformed by AI buys from someone else.

In the United States, where these habits are best measured, the numbers are already in. Visitors sent by AI are more numerous every month, they buy more than others, and they follow what the AI tells them.

×13

The rise in visits sent by AI to US retail sites between October 2024 and July 2026.

Adobe Analytics
+60%

A visitor sent by AI buys 60% more often than another visitor, and spends 53% more per visit.

Adobe Analytics
79%

of shoppers who use AI have already switched brands because it recommended another one.

Envision Horizons
1 in 2

has already abandoned a purchase after the AI raised a concern about the product.

Envision Horizons
1 product in 3

In our own test on 50 products from US brands, the brand's site never appeared in the answers for 16 of them. The AI answered from sites other than the brand's own.

1 traveller in 3

in the US already plans trips with an AI such as ChatGPT. That is five times more than in 2024.

These customers arrive ready to buy, and they go where the AI sends them. If your site gives it wrong or incomplete information, it sends them elsewhere.

Check my site for free

Sources: Adobe Analytics, July 2026 data on more than 1 trillion visits to US retail sites (Digital Commerce 360, 19 August 2026). Envision Horizons, February 2026 survey of 1,035 US adults (eMarketer). Phocuswright, “The AI Surge”, 2026. GetRaight test, November 2026. These figures describe the US market; they do not predict your site's revenue.

See what ChatGPT says about you.

Free, no sign-up: your score in 1 to 2 minutes, then ChatGPT's and Gemini's answer about one of your products, compared with your site.

Analyse my site for free