AI Search: Which Sources to Prioritize? The Level-by-Level View
Summarize this article with AI
I look at 15 sites per week. They all have the same problem.
I look at 15 sites per week. They all have the same problem.
They track 4 sources. AI exploits dozens.
Google Search Console, a few backlinks, keyword rankings. That’s the historical reflex. But conversational chatbots, generative answers, AI agents rely on sources these tools don’t show.
Chris Green, technical consultant at Torque Partnership, published on Search Engine Journal a hierarchy of data sources for AI Search. His observation is blunt: teams suffer from source myopia. They look at what they’ve always looked at, while AI reads elsewhere.
He classifies sources into 4 levels. Tier 1: confirmed and current, used for grounding and actions. Tier 2: confirmed and current, used for training and licenses. Tier 3: confirmed historical, used for pre-training. Tier 4: probable, not officially confirmed.
For e-commerce, this hierarchy shows where to invest first. Skip tracking 47 sources. Track 6.
The problem is simple. Concrete.
I’ll show you how. How many sources do you track?
One e-commerce client invested $12 000 in the wrong place
An e-commerce client selling electronic spare parts calls me on a Tuesday morning. He put $12 000 into buying guides and long-form articles. His organic traffic is flat. His appearances in ChatGPT? Almost zero.
The diagnosis comes fast. The content isn’t the issue. The problem lies in the source architecture.
His product pages aren’t in Google Merchant Center. His OpenAI product feed doesn’t exist. His Google Business Profile listings are incomplete. Meanwhile, his competitors organize their data cleanly and show up in AI answers.
We prioritized by levels. First Tier 1: Merchant Center, Business Profile, product page optimization for Google and Bing grounding. Then Tier 2: a product feed for OpenAI, CSV format, refreshed every 15 minutes.
In 3 months, citations in AI answers went from 3 to 47 mentions across 22 target queries. That’s what I saw with this client. Direct/other traffic, which often captures ChatGPT callbacks, climbed 12 %.
Why? Real-time AI answers rely on Google and Bing grounding, and on product feeds. Not on blog posts from 2024.
Result. Mechanism. No ads.
The Level-by-Level View: Tier 1 to Tier 4, from Search Engine Journal
Chris Green’s hierarchy, published on Search Engine Journal, classifies sources into four levels. His logic: the degree of confirmation, and the current use AI makes of them.
Tier 1 — Confirmed and current, used for grounding and actions.
- Google Search: grounding connects Gemini to the live web, with inline citations.
- Bing Search: Microsoft documents the use of Bing results to improve Copilot’s answers.
- Google Merchant Center: the product feed feeds Google’s shopping surfaces.
- Google Maps: grounding with Maps brings geospatial context.
- Google Business Profile: profile data feeds local surfaces.
These sources are actionable. You can optimize them today. They directly influence AI answers rich in citations.
Tier 2 — Confirmed and current, used for training and licenses.
- OpenAI product feeds: merchants submit a secure CSV/JSON feed, refreshed every 15 minutes. Inside: identifiers, descriptions, prices, stock, media, shipping. ChatGPT therefore displays products with precision.
This level is recent, but it’s already running. For e-commerce, it’s an immediate lever.
Tier 3 — Confirmed historical, used for pre-training.
- Common Crawl: GPT-3 used about 60 % of filtered Common Crawl in its sampling mix. LLaMA 1 reported 67 %.
- Historical web corpora (C4, etc.): C4 is a cleaned derivative of Common Crawl. LLaMA reported C4 at 15 % of its pre-training mix.
You don’t optimize these sources directly. They show a past state of the web. Working on them is long-term.
Tier 4 — High probability, not officially confirmed.
- Microsoft Merchant Center: commercial infrastructure parallel to Google Merchant Center, inferred.
- Marketplace feeds: Shopify is already integrated into ChatGPT, which supports the pattern.
- Third-party grounding services: retrieval intermediaries, with no single canonical source.
Worth watching, but that’s not where to invest first.
This hierarchy directly applies the DOSE framework, taught by Guillaume Attias (BMO Academy): clarify, order, simplify, execute. It lightens mental load. You stop scattering budget.
Tier 1 and Tier 2: Invest Where It Really Counts
For e-commerce, everything starts with Google Merchant Center. An up-to-date feed, with GTIN, price, stock, images, feeds shopping surfaces and AI product answers. A misconfigured feed, with rejected products or missing attributes, makes you invisible.
Google Business Profile does the same job for local. AI answers that compare stores or recommend a point of sale rely on these listings. An incomplete listing, without up-to-date hours, without secondary catégories, loses citations.
The OpenAI feed is the channel that just arrived. OpenAI documents secure feeds, updated every 15 minutes. Products appear in ChatGPT Search and future shopping interfaces. It’s an immediate lever, still underused.
The classic mistake: neglecting these feeds because they don’t generate direct clicks visible in Search Console. But they weigh on citations. A product listing with the right attributes is more likely to be cited in a comparative answer. A competitor without a feed stays invisible.
Google and Bing grounding: every product page must answer factual questions. Price, availability, compatibility, dimensions. AI answers extract passages from web pages. If your page doesn’t contain the answer, another one will be cited.
Why does it work? The brain works by hierarchies. The DOSE framework (Guillaume Attias, BMO Academy) teaches reducing cognitive load by ranking priorities. Three levels, not forty.
Nothing complicated.
Avant de parler de contenu, regardons ce que l’IA ingère à la base. Ces chiffres viennent de Search Engine Journal et ils changent l’ordre des priorités.
Où l’IA puise ses données d’entraînement
Common Crawl écrase tout : le pré-entraînement n’est pas votre terrain de jeu.
Long-form content is no longer enough: structured data is the new lever
Most teams invest in long-form content. It’s reassuring. But it’s Tier 3.
AI Search doesn’t stop at pre-training. It queries the web in real time via Google and Bing. It reads product feeds. It checks local listings. A blog post from 2023, even excellent, weighs less than a structured, up-to-date product listing accessible via Merchant Center.
Chris Green says it on Search Engine Journal: chatbots draw from more and more sources. The job gets harder. And more interesting.
In my opinion, the next SEO battle is fought outside rankings. It’s fought over data sources. A site with 800 references and a clean feed will beat a site with 800 articles and no feed.
The reason: transactional AI answers rely on product data. Buying guides feed content. Product listings trigger the citation. Grounding fetches the exact data, prose doesn’t interest it.
How many sources do your teams actually track?
Often four. Sometimes three.
Meanwhile, AI consults dozens.
Pas besoin de jargon. Trois étapes suffisent pour attaquer vos sources IA dans le bon sens, du Tier 1 vers le reste.
Trois étapes pour prioriser vos sources
Un plan simple pour remettre votre architecture de données dans le bon ordre.
Three Concrete Steps to Prioritize Your Sources
The action plan. Simple, no jargon.
Step 1: audit Tier 1 sources.
- Google Merchant Center: check errors, rejected products, missing images. A feed with 12 % rejected products means citations go elsewhere.
- Google Business Profile: complete hours, catégories, attributes. Add recent photos.
- Bing Places: often forgotten, but Copilot relies on Bing Search.
Step 2: prepare a product feed for OpenAI.
- Use the CSV or JSON format documented by OpenAI. Include identifiers, descriptions, prices, stock, media, shipping times.
- Configure an update every 15 minutes. Stale data erodes trust.
Step 3: optimize product pages for grounding.
- Implement Schema.org Product, Offer, AggregateRating.
- Answer factual questions at the top of the page: price, availability, compatibility, warranty.
- Write concise descriptions, with technical attributes highlighted.
What I see with my e-commerce clients: Merchant Center is configured, but full of silent errors. Rejected products, non-compliant images, outdated prices. Fix that, and citations come back.
This priority is DOSE (Guillaume Attias, BMO Academy). Three high-impact sources, and mental load drops.
Nothing complicated. Just execution.
Measure, Iterate, and Stop Chasing Everything
Measuring the impact of AI sources means going beyond Search Console. ChatGPT callbacks often land in the « direct/other » bucket. Watch variations in that traffic. Run regular test queries: « best product X for Y », « X vs Z comparison ». Note by hand if your brand is cited.
Let’s go back to the e-commerce client. After 4 months, AI citations went from 3 to 47 mentions across 22 target queries. Direct/other traffic climbed 12 %. An observed order of magnitude. No figure from a magic tool. Manual tracking, week after week.
Iterate. Tier 4 may become Tier 1 tomorrow. Microsoft Merchant Center, marketplace feeds, third-party grounding services: keep an eye on official announcements. Adjust.
Why does ranking work? The human brain avoids cognitive overload. Dozens of sources to track paralyze. Three clear levels trigger execution. That’s the foundation of Guillaume Attias’s DOSE framework (BMO Academy): reduce the load to decide faster.
You don’t need to master everything. You win by mastering the six sources that matter.
And you, how many sources do you really track?
AI Search Audit: Identify Your 3 Priority Sources
I show you the pages to structure and the feeds to activate first. A live audit, no sales pitch.
Book a strategic call — 45 minFrequently Asked Questions
Which AI Search sources are most important for e-commerce?
Google Merchant Center, Google Business Profile, OpenAI product feeds, and grounding via Google Search and Bing Search. Those are my Tier 1 and Tier 2 sources. I can activate them right away.
Does Google Merchant Center influence ChatGPT’s answers?
Yes, thanks to grounding. ChatGPT cites products whose data is structured and up to date in Google Merchant Center. OpenAI also accepts a dedicated product feed, refreshed every 15 minutes.
Should you optimize for Common Crawl or C4?
No, not as a priority. These sources serve pre-training (Tier 3), with mostly historical value. Nothing directly actionable. Your time is better invested in feeds and structured content that engines consult in real time.
How do you measure the impact of AI sources on your traffic?
Watch direct/other traffic. Most ChatGPT callbacks come through there. Manually track brand mentions on your target queries: these citations, classic tools don’t capture them yet.
Where do you start an AI Search strategy for an e-commerce site?
I start with a Google Merchant Center audit. I complete Google Business Profile. I fix feed errors. Then, a feed for OpenAI. And on product pages: Schema.org, plus factual answers.
