We know that AI assistants like chatgpt access search indexes (such as Google and bing) can retrieve the response of URLs. But how, what?
To find out, we conducted a series of experiments to see the relationships between URLs referenced by the AI assistant and the results found in Google when searching for the same topic.
So far, we’ve tested it Long tail tips (Like the ones you want to go into chatgpt, very long, very specific queries); Fans ask (Medium-length hint related to the original long-tail hint); today, we are testing the short-tail keyword – Urtra short and specific “head” term.
The short-tail keyword provides the clearest illustration of how AI citations can be tracked through Google results.
Based on three independent studies, we concluded that ChatGpt (and similar systems) not only extract URLs directly from Google, Bing, or other indexes. Instead, they applied additional processing steps before citing the source.
Even if we checked fan-eliminated queries (actual search prompts these systems to be sent to search engines), the overlap between AI and search engine references is surprising.
In other words, while Chatgpt may draw inspiration from Google’s search index, it seems that its own selection layer is still applied to filter and shuffle links to appear.
So this is not enough to identify fans’ queries and rank them high, which are other factors that influence the surface of URLs, which are outside the publisher’s control.
Different query types tell us different information about how AI assistants process information.
In our earlier study, data scientists at Ahrefs xibeijia guan Analytical citation overlap between AI and search results to obtain information Long tail and fan Tip, use Ahrefs brand radar.
This time, she took 3,311 classic SEO-style head samples covering information, business, transactions and navigation intent.
| Sample query | Information | Commercial | trade | navigation |
|---|---|---|---|---|
| 1 | Cincinnati Panda Basketball | Best Credit Card Rewards | Pool for sale | OneDrive signature |
| 2 | Protein in shrimp | TV audio bar | Shop girl’s dress | Verizon Customer Support |
| 3 | What is cybersecurity | Sauna at home | Buy a domain | Costco toilet paper |
Each keyword is run through chatgpt, confusion and Google’s top 100 SERPs to analyze reference overlap between AI and search.
If there is anything that can closely align with Google’s results, then you want it to be a short-tail question, as this is the classic way we search.
But that’s not the case.
And the references overlap Short-tail query (10%) Fans ask (6.82%), it is still much weaker than expected directly to echo SERP.
This is even more surprising, now we have confirmed Openai and confusion keep scratching Google Results through third-party providers.
If our research focuses only on “real-time” queries (e.g., news, sports, finance), we may see more overlap It is said that These types will crawl Google.
Confused quotes closely match Google search results for short-tail queries.
Unlike chatgpt, overlap is not only visible at the domain level, but also, most of the confusing reference pages are also the exact URL rankings in Google’s top 10.
This reflects our findings Long tail query Research, the confusing response is most similar to Google’s results, thus strengthening its design as a “citation first” engine.
Domain overlap is always higher than URL overlap, which suggests that Chatgpt and confusion reference the same website as Google, but not exactly the same.
In Chatgpt, the domain-URL gap is particularly wide – 31.8% vs. 10%.
In other words, Chatgpt references a ranking domain that is about 3 times higher than the ranking page.
On the one hand, this may mean chatgpt selection Different page Same The domain name is Google.
For example, Google quotes a page ahrefs.com/writing-tools/and chatgpt found a better “fit” ahrefs.com/blog/ And quote another.
If true, this enhances the value of creating clustered content – push multiple pages of multiple pages into different theme intents in order to find the best chance.
Another possibility is that they all depend on the same pool authoritative Domain, but disagree on any page.
Evaluate your cluster content in AI and search
You can view the SEO performance of cluster content in the relevant terminology report in AHREFS keyword Explorer.
This will show you if and Where You can span the entire group of related keywords.
Simply add the parent topic filter, and a target filter that contains your domain.
After completing this, go to Ahrefs brand radar Check the AI performance of cluster content.
Run a single URL through referenced page reports Ahrefs brand radar See if your cluster content is referenced by AI assistants such as Chatgpt, Confused, Gemini, and Copilot.
Figuring out if anything is missing on either surface, optimize until these gaps are filled and the entire cluster is enriched.
You can use topic gap suggestions ahrefs’ AI Content Assistant Help with this.
Short-tail queries show closer SERP-ai alignment than natural language prompts, especially when confused.
But the chatgpt reference generated by the fan elimination query (first Learned by SQ and Xibeijia) Shows minimal overlap. They only match 6.82% of Google’s top 10 results.
We don’t compare Apple here. These percentages represent different studies and data sets of different sizes.
However, each study produced a similar finding: the pages cited by Chatgpt do not significantly overlap with those ranked by Google. This is very much the opposite.
Another thing we haven’t mentioned yet is intention. The larger citation overlap we see in short-tail queries can be explained in part by the relative stability of navigation, commercial and transaction queries, which we did not evaluate in previous studies.
The term SERPs for navigation, business and transaction headers tend not to be too frequent, as the collection of related products, brands or destinations is limited.
This stability means that AI Assistant and Google are more likely to converge on the same source, which means that the amount of overlap is higher than the information query (where possible page pools are much larger and more volatile).
The final thought
In all three studies, the story was consistent: Chatgpt does not follow Google’s source, which is confusing.
Surprisingly, Chatgpt is very different from Google when we know Openai Do Scratch Google results.
My intuition is that Chatgpt is more than just confusing about distinguishing its results from Google.
This theory comes from Square footage It seems to be the most likely to me:
“ChatGpt may use a hybrid approach to get search results from various sources, such as Google SERP, Bing SERP, their own indexes and third-party search APIs, then combine all URLs and apply their own rearrangement algorithm.”
Regardless, search and AI are shaping discovery side by side, and the best strategy is to build content that gives you a chance to appear on both surfaces.
