AI Sources: The Hierarchy Beyond the Data SEOs Track
Summarize this article with AI
SEO tools only see a fraction of AI sources
A client calls me on a Tuesday morning. He invested $8,000 in content production. 47 articles. A well-built semantic cocoon. His site is clean, fast, well indexed.
And yet.
His name appears nowhere in ChatGPT, Perplexity, or Google AI Overviews responses.
Zero citations. Zero clicks.
The problem came from the sources, not the content.
I look at 15 sites per week. All have decent Google rankings. Clean Search Console reports. SEO tool suites showing upward curves.
And meanwhile, AI models cite other brands. Other product listings. Other local pages.
Why? Classic SEO tools only track a fraction of the sources AI actually uses. According to a grid published by Chris Green on Search Engine Journal, there are at least four levels of data sources. Most SEO teams stop at the first.
Result: we optimize the visible. And leave behind what triggers citations.
The question that matters isn’t « Am I visible in Google? »
It is: « Which sources do AI models pull their answers from, and which ones am I really monitoring? »
Avant d’entrer dans le détail, posons la grille en un coup d’œil. Quatre niveaux de preuve, du plus fiable au plus incertain, qui déterminent où votre marque doit vraiment être présente.
La hiérarchie des 4 niveaux de sources IA
Du grounding temps réel aux hypothèses : la grille SEJ qui classe les sources que les modèles citent.
Four levels of evidence, not fifty sources to churn through
When you discover the scale of the problem, the temptation is to track everything. Every platform. Every feed. Every corpus.
Bad idea.
Search Engine Journal published a grid that ranks sources by evidence, not by volume. Four levels.
| Level | Typical use | Examples | Evidence status |
|---|---|---|---|
| Tier 1 | Grounding / real-time actions | Google Search, Bing Search, Google Merchant Center, Google Maps, Google Business Profile | Confirmed + current |
| Tier 2 | Training / licensing | OpenAI product feeds | Confirmed + current |
| Tier 3 | Historical pre-training | Common Crawl, C4 | Historically confirmed |
| Tier 4 | Third-party grounding services, marketplace feeds | Microsoft Merchant Center, retrieval intermediaries | Strong evidence / highly probable |
Tier 1: the sources that power real-time answers, with direct citations. Tiers 2 and 3 influence the model’s knowledge, less directly. Tier 4 is probable, not yet confirmed.
This grid changes the order of priorities. Instead of chasing fifty weak signals, you put effort where the evidence is established.
In short: start with Tier 1. Only then, move down.
Tier 1: being present in grounding, not just the index
Grounding is a model’s ability to fetch real-time content to support its answer. Search Engine Journal confirms it: Google Search serves as a grounding source for Gemini. Google results are plugged into the model, with citations pointing back to source URLs.
It’s the same for Copilot with Bing Search. Microsoft documents it: Bing results feed into responses.
But there’s something more unsettling.
Google Merchant Center is a documented Tier 1 source. Product feed data powers Google shopping pages. A poorly filled product listing won’t surface in AI responses about that product.
Google Maps is also a documented grounding source. Models pull geospatial context from it.
And Google Business Profile is almost never discussed in SEO. This listing drives local results. I saw a hotel client in Southeast Asia get cited in Google AI Overview for a local query. It wasn’t his site being cited. It was his Google Business Profile listing.
The lesson is brutal: you can have the best site in the world, but if your local listing, Merchant feed, or Maps data aren’t up to date, you stay invisible in AI responses.
Who in your SEO team still monitors Google Merchant Center?
Common Crawl, C4, and product feeds: the hidden side
Tiers 2 and 3 work upstream, during model training. They don’t cite anything in real time. Their role: deciding what the model retains about your sector.
Common Crawl is a massive corpus of web pages. Search Engine Journal published the grid: GPT-3 used about 60% filtered Common Crawl in its sampling mix. LLaMA 1 reported 67%.
C4, a cleaned version of Common Crawl, accounted for 15% of LLaMA’s pre-training mix. These numbers matter. Your pages in these corpora feed the models’ général knowledge.
Careful though: appearing in Common Crawl doesn’t guarantee any citation. Models don’t query Common Crawl in real time; they digested it during training. The citation itself often comes from grounding.
Tier 2 also includes OpenAI product feeds. Merchants share a secure CSV or JSON feed, refreshed every 15 minutes, with identifiers, descriptions, prices, inventory, media, and delivery options. ChatGPT then displays accurate, up-to-date products.
That’s huge.
A well-structured catalog, sent as a feed, gives an AI what it needs to cite your products without going through your site. Most SEOs completely ignore this visibility source.
Tier 3 remains a matter of history. A young or poorly interlinked site has little chance of being well represented in Common Crawl. It takes months. And it builds.
Blind spots: products, local, assistants
SEO teams spend hours on Google rankings, tags, internal linking. Meanwhile, three blind spots keep them away from AI citations.
First blind spot: product feeds. According to the SEJ source, Shopify is already integrated into ChatGPT for shopping. Shopify catalog data flows directly into responses. If you sell on Shopify, your products may already be cited. Or not, if your catalog is poorly filled.
Second blind spot: local data. Google Maps and Google Business Profile are confirmed grounding sources. An incomplete Google Business Profile listing, an imprecise address, outdated hours: you lose local citations. Among my clients, I’ve seen well-filled local listings get more mentions in AI Overviews than pages optimized for classic SEO.
Third blind spot: assistants and agents. ChatGPT, Perplexity, Copilot. These tools don’t settle for Google. They cross-reference sources. If your presence is limited to your site and Google Search Console, you’re absent from part of the spectrum.
Tier 4, which includes Microsoft Merchant Center and third-party grounding intermediaries, remains on the watchlist. The evidence is strong, but not yet confirmed. It shouldn’t come before Tiers 1 and 2.
Prioritizing means making a choice. And it starts by looking at what we don’t see.
Prioritize by evidence level, not data volume
Since I’ve been applying this grid with my clients, audits have changed shape. We no longer start with Google rankings. We start with source mapping.
My DOSE method, built with Guillaume Attias at BMO Academy, has four steps: Diagnosis, Objectives, Strategy, Execution. Source diagnosis comes first.
Concretely, here’s what I do.
- List all sources where the brand appears or should appear: site, local listings, product feeds, marketplace catalogs.
- Classify each source according to the four levels of the SEJ grid.
- Handle Tier 1 first: Google Search, Bing, Google Merchant Center, Google Maps, Google Business Profile.
- Add OpenAI product feeds to Tier 2 with a 15-minute refresh frequency.
- Check presence in Common Crawl and C4 for the long term, without making it a short-term priority.
Then we measure AI citations, not just rankings. We observe which sources are cited. We fix those that aren’t.
The result is often counterintuitive. One client saw his AI citations climb after updating his Google Merchant Center feed, while his classic organic traffic hadn’t moved. No new page. No new article. Just a properly fed source.
The first client call, I do it as a live audit. I don’t sell the method. I show the pages.
So, which source that you aren’t tracking yet will your next AI citation come from?
AI Audit: I map your actually cited sources
During the live audit, I run your data through the 4 levels. You leave with a prioritized list of sources to fix or feed, without sorting through fifty useless spreadsheets.
Book a strategic call — 45 minFrequently Asked Questions
Which sources do AI systems use most to cite a site?
AI starts from sources it can verify: Google Search, Bing Search, Google Maps, Google Business Profile, Google Merchant Center. Then come product feeds and training corpora, with Common Crawl at the top.
Is being present in Common Crawl enough to be cited by ChatGPT?
No. Common Crawl is used for pre-training, not grounding. A page can appear there without being cited once. Bet first on sources that models read in real time.
How to appear in Google AI Overviews?
The most direct way: be present in grounding. A page that ranks well in Google Search, an up-to-date Google Business Profile listing, a clean Google Merchant Center feed. That’s where citations come from.
Do product feeds help get AI citations?
Yes. OpenAI accepts CSV/JSON product feeds refreshed every 15 minutes. Shopify catalog data is already in ChatGPT. These feeds help assistants cite your products, prices, and availability accurately.
Which tools track these AI sources?
Few tools cover the entire field. Classic SEO suites track Google and Bing, never product feeds or Google Maps. So I do the audit manually, level by level: I list the sources, then check their presence and freshness.
