Perplexity Sonar
« Comment surveiller la pression d'huile sur une BMW E30 ? »
Sources read by the AI
- 6enligne.net
- reddit.com
- youtube.com
- mecatechnic.com
- cyl6.com
Four of the five sources are community spaces. Only one is a shop.
E-commerce
Does your store exist when the AI recommends a site?
A buyer asking "where can I buy this" no longer gets ten links, they get three names. We measured two complete e-commerce markets — and they behave nothing alike.
Report within 48h. Secure payment by Stripe. No commitment.
- 40
- real questions
- 3
- AI assistants
- 3
- times each question is asked
- 360
- measurements in total
The study 1/2
One market measured: automotive spare parts
40 buyer questions put to 3 AI assistants, three times each — 360 answers analysed on a real e-commerce market. We counted every brand named in the body of the answers, not just the sites read as sources.
The brands AIs recommend in this market
360 answers · confirmed brands- 1driftshop.com29
- 2jrparts.fr27
- 3mecatechnic.com24
- 4schmiedmann.com22
- 5ecstuning.com21
- 6myclassicparts.com21
- 7auto-doc.fr21
- 8etsy.com18
The best-placed site in this market is named in just 8.1% of answers. The runner-up, 7.5%. In other words: no store is established in AI answers on this market. The ground is open — for now.
ChatGPT
« Kit jauge température d'huile pour BMW E28 ou E34, quelles marques existent ? »
Sources read by the AI
- vdo-instruments.com
- renditioncustom.com
- prosportgauges.com
- lewonze.de
Four manufacturers cited, three of them foreign. No French reseller in the answer.
Claude Sonnet 4.5
« Où acheter des pièces détachées rares pour BMW E30 en France ? »
Sources read by the AI
- piecesauto24.com
- driftshop.com
- pafclassic.com
- pieces-bmw.fr
- ovoko.fr
- lesanciennes.com
Six online stores for a single question. Across the 360 answers, none exceeds 8%.
The study 2/2
A second market: specialty coffee
10 buyer questions put to the same 3 assistants, three times each — 90 answers, in French, English and Spanish. Same method as above, radically different market.
The brands AIs recommend
90 answers · August 2026- 1Terres de Café28
- 2L'Arbre à Café12
- 3Baranca Coffee8
- 4Belco7
- 5Kawa6
- 6Lomi6
- 7MaxiCoffee6
One roaster is named in 31% of answers — four times the leader of the previous market. Same method, same sector, inverted structure: here the ground is taken, and it has to be fought for.
Perplexity Sonar
« Où acheter du café de spécialité en ligne en France ? »
Sources read by the AI
- terresdecafe.com
- reddit.com
- graindecafe.com
- brulerie-dalre-shop.com
- plainecafe.com
Four shops out of five sources. Here the AI reads merchants, not forums.
ChatGPT
« Where can I buy Colombian specialty coffee online in Europe? »
Sources read by the AI
- caferaizal.es
- barancacoffee.nl
- unumacolombia.com
- sierranevadacaffe.com
- delcielofarm.com
Five European roasters, not one French. Asking in English changes the whole ranking.
Claude Sonnet 4.5
« Quelles marques de café achètent vraiment en direct aux producteurs ? »
Sources read by the AI
- malongo.com
- cyrilrouquet.com
- larbreacafe.com
- atelierbraam.com
- lepanieracafe.fr
On the ethics question, the brands' own websites are what counts.
Two markets, two opposite strategies
The contrast between our two measurements is the most useful result on this page — and it rules out any one-size-fits-all recipe.
In spare parts, AIs read communities. Specialist forums, Reddit threads, video channels: four community sources for every shop. No merchant exceeds 8% of mentions there. The ground is open, and it is won by existing where the conversation happens.
In coffee, AIs read the merchants themselves. Roasters' own sites dominate the sources, only one forum appears, and a single player holds 31% of mentions. The ground is taken: it has to be fought for, by working on what makes a site quotable — a sharp speciality, verifiable claims, an identity a writer can sum up in one sentence.
A third lesson, visible in both markets: asking the question in English changes the entire ranking. If you sell beyond your home country, measuring in one language only gives you a false picture of where you stand.
What the AI needs to find on your product page to recommend you
Being named in an answer doesn't mean the AI actually read your site properly. An agent that has to recommend a place to buy — not just cite a brand — needs unambiguous information: current price, real availability, delivery time, return conditions. That information has to exist in a format a machine can read directly, not just inside an image or a description paragraph.
Structured data (schema.org Product and Offer) on every page
An explicit price and currency
An up-to-date stock status
Pages that aren't blocked to the crawlers reading on an assistant's behalf
A page that's ambiguous on those points isn't bumped to second choice — it's simply dropped in favour of a source that answers more cleanly, often a marketplace whose entire job is aggregating exactly that information.
Checked on 500 real French e-commerce sites, 452 reachable for analysis:
82%
have no usable schema.org Product or Offer data at all
14%
publish a complete page — price, currency and stock a machine can read
Many of these sites rank well on Google. It's a blind spot no standard SEO audit covers, because it isn't what Google looks at first — but it's exactly what a purchase agent looks at before recommending you. In practice, these sites drop out of the answer the moment a buyer asks a precise question, replaced by a competitor whose page answers unambiguously, or by a marketplace that aggregates exactly that information.
How we measure
No question contains your name. That is the condition for the result to mean anything: we look at whether the AI mentions you spontaneously, not whether it can describe you when prompted to.
- 01
40 questions written for your market
Discovery, comparison, best choice, alternative to a competitor, local search, purchase decision: the 40 questions cover the intents a buyer actually moves through. They are written from your business, your site and your area — never generic, never recycled from one client to the next.
- 02
3 AIs asked, 3 times each
Perplexity, ChatGPT and Claude, with web search enabled. Each question is asked three times, because an AI never answers the same way twice: a single measurement proves nothing.
- 03
360 answers analysed
We count your citations, your recommendations, where they sit in the answer, and every domain the AIs went to read to form an opinion.
The assistants we query
Three engines, all with web search enabled — under the real conditions in which your buyers use them.
Perplexity
Sonar
Built-in web search. The most source-driven engine of the three.
ChatGPT
GPT, consumer model
The model actually served inside ChatGPT, not an API variant.
Claude
Sonnet 4.5
Native web search, long and highly structured answers.
What you get
A report you read in fifteen minutes and turn into decisions.
Your citation rate
Out of 360 answers, how often your site is a source the AI actually read. Broken down per assistant.
Your recommendation rate
How often your brand is named in the answer, and in which position — being cited first is not worth the same as being cited last.
The competitors taking your place
A ranking of the brands AIs recommend instead of you. Not the ones you think you have: the ones the data points to.
The sources AIs read
The domains cited most often in your market — forums, comparison sites, trade media. That is where your presence is decided.
The questions where you are absent
The question-by-question detail, so you can see exactly which topics you do not exist on.
Where to start
The three or four levers your data points to: which domains to get cited on, which topics you are absent from, what is holding your site back. Derived from the measurement, not from general principles.
Two plans
The snapshot, or the film. No commitment either way.
The Audit
€149excl. VAT
one-off
Where you stand today.
- 40 questions written for your market
- 3 AI assistants, each question asked 3 times
- Your citation and recommendation rate
- The competitors taking your place
- The sources the AIs read in your market
- A technical check: what, on your site, stops an AI from reading it properly
- Where to start: 3-4 levers drawn from your data
- PDF report within 48h, invoice with VAT
The Monitoring
Recommended€59excl. VAT/mo
cancel any time
First month €149 excl. VAT — your full audit — then €59 excl. VAT per month.
Your visibility, watched month after month.
- Everything the Audit contains
- Mentions gained and lost since last month
- The competitors that overtook you, and on which questions
- New sources that started citing you — or ignoring you
- This month's fixes, in order of impact
- Cancel any time, on request
Email us to set it up
Set up by hand for now: write to us and we reply within 24h.
Report delivered within 48 business hours — fully refunded if we run significantly late.
What next?
Would you rather the fixes were made for you? Reply to your report and we will talk it through. Quoted case by case, from what your own figures point to.
Start your audit
Five fields. They are what we use to write your 40 questions — the more precise you are, the more accurate the audit.
Frequently asked questions
See another industry
Hotels
"Where to stay near Mont-Saint-Michel" is exactly the kind of question people no longer put to Google. We measured what the AIs answer, across 90 real queries.
Campsites
"Which campsite for a stay in Normandy" is exactly the kind of question people now put to an AI rather than to a search engine. We measured what it answers.
All industries
The method, the numbers and the price, without an industry angle.