In the 15,000 hint dataset, we found that on average only 12% of the links, with links referenced by Chatgpt, Gemini and Copilot appearing in the top 10 results on Google, appearing in the same hint.
Confusion is outliers: nearly one-third of its citations point to the top ten pages in the target query.
We analyzed four AI assistants, namely Chatgpt, Gemini, Copilot, and Confusion, to measure the frequency of the URL they reference or the frequency of overlapping Google or Bing levels for the same query as Google or Bing.
On average, AI Assistant overlaps 11% of citations between the top 10 of Google and Bing.
((Google AI overlap total + Bing AI overlap total) / 10) = 10.96%
In addition, we take a look at the single search engine breakdown.
Use from Ahrefs brand radarour data scientist, xibeijia guansearched Google and Bing for 15,000 long-tail queries, and then asked the same question to the AI assistant.
Example queries include:
- “How much does it cost to install a security camera?”
- “How do you get enough protein when you can’t eat dairy?”
- “What is the cheapest comprehensive insurance?”
- “�Cuálesel Mejor suplemento para larecuperación muscles?”
Xibeijia then extracts all visible references returned by each question in the search and AI index.
In the search dataset, she recorded the ranking positions of the referenced URLs and then classified them based on whether they appear in the top 10 or top 100 results.
Tested AI assistants include CHATGPT (text citation and response bottom reference), Gemini, copilot and confusion.
Based on research conducted by our data scientists xibeijia guanOnly 12% of citations in AI Assistant also ranked on average among Google’s top 10.
((28.6 + 8 + 6.1 + 8.6 + 8.2) / 5) = 11.9%
80% of the original queries in these references are not ranked anywhere in Google – 80% are essentially the average of all the “red” columns in the picture above.
The overlap between AI and search citations is smaller than we expected, but we have several theories that might be.
We’ll introduce them later.
Confusingness is consistent with Google’s recent
Confused is an AI assistant, actually Built as a quote– Its goal is to support almost every statement of one source.
Confusion is the world’s first answer engine. It searches the internet in real time to provide fast, clear answers to any question, including sources and citations… Each answer comes with clickable citations, making it easy to verify information and dig deeper if needed.
Confused teamOctober 2024
Chatgpt and Gemini, on the other hand, are not always linked.
Their citation process is very queries-dependent, and their internal systems trigger searches based on different criteria.
For example, chatgpt seems to use something called Sound Classifier– If the prompt “probability” exceeds a certain threshold (i.e., if the response is not easy to predict based on training data), ChatGpt will perform a web search.
Confusion always takes advantage of content that ranks well in Google, with 28.6% of cited URLs landing in the top 10.
For other AI assistants, that number hovers around 8% – more than 80% of references come from pages that don’t have target queries at all.
Since confusion focuses on citations, its results will be closely aligned with Google’s results, which seems logical.
But unlike Gemini and other AI assistants, confusion doesn’t attract Google or Bing’s index. In fact, it has a search index based on its tracks: Perplexitybot.
AI assistants (even those who borrow from Google and Bing) query search indexes in fundamentally different ways. This explains why we see the smallest crossover.
When measuring for BING results, overlap remained low in the top 10, with an average hit of about 10%.
((16.6 + 14 + 8.1 + 8.1 + 3.3) / 5) = 10.02%
Copilot rises to the top of the list, which is expected because it is a Microsoft-owned product Use Bing to search results when it generates its response.
Confusion is still one of the most unified AI assistants, with Chatgpt and Gemini having the least overlap.
This study is based on our previous study that advanced pages are cited in Google’s AI overview.
Further reading:
In these studies, we found that there was a moderate positive correlation between Google ranking and citations in the AI overview, with 76% of cited URLs coming from the top ten pages.
Based on data, it seems that the capabilities outlined by AI are like extensions of traditional search, which mainly reflects SERP.
In other words, the AI overview follows SERPS – the AI assistant does not.
Even the assistant built it go through Google and Google’s SERP are not close.
Obviously there are other factors that affect the content cited, not just search rankings.
Our data show that the URLs referenced by AI assistants rarely match Google and Bing levels.
In most cases, for target cues, they don’t appear in the top 10 – usually, they are not in the top 100 at all.
This is because AI assistants do not rank results in the same way as search engines.
They tend to retrieve pages based on multiple changes in that query, rather than just processing a user query, but a kind of “Query fans. ”
These changes are then merged using a similar approach Mutual level integration ((RRF), which seems to be always favored.
For example, the pages on “How to Lower Coffee Machines,” “Cleaning Nespresso Machines” and “Removing Limesscale from Coffee Machines” are ranked sixth on page 6 and may be ranked first in only one of them.
In fact, AI assistants may never look for Actual Prompt the user to enter – it is just a catalyst for other fan phase-out query.
This is why the final quotation is not always ranked for the original prompt.
Personalization can also work: making the exact same search query through AI can lead to completely different references due to factors such as user prompts or conversation history.
All of this explains why AI references do not always reflect SERPS.
hint
You can use parent topic to optimize in multiple long-tail query variants AHREFS keyword explorer.
Simply search for a keyword, go to the match or related term report, and view the parent topic on the left, or hit the cluster via the Parent Topics tab.
Then click on the “Problem” tab to get the long tail query to locate your content…
To see how much ownership you have in your existing long-tail query permutation, add a target filter for your domain.
Recent tests aleksis rylko It shows that search citations are very similar to Google citations.
Rylko believes Chatgpt may now quietly get from Google’s search index.
Aley da Solís confirmed this– She found that Chatgpt uses Google as a fallback when the Bing page is out of reach.
These findings surprised many people because Chatgpt’s built-in provider is Bing.
From our own dataset, we don’t see much difference on how Google and bing results surfaces surfaces.
| Google (Top 10) | Bing (Top 10) | Google (Top 100) | Bing (Top 100) | |
|---|---|---|---|---|
| chatgpt (text quotation) | 8.00% | 8.10% | 8.80% | 5.00% |
| chatgpt (reference) | 6.10% | 8.10% | 9.20% | 5.20% |
That said, we collected this data in early July, so the results may have changed – especially given the Openai plan to release its AI-First Web BrowserIt is even possible to establish your own search index Give way.
We will conduct further research to study any citation changes.
Summarize
From our research, ranking well does seem to help, but it is not necessarily a query you might think you need a well-ranked one.
We will continue to look at the overlap of citations in traditional search and AI to see if there are any changes.
Meanwhile, building theme clusters and “own entities” optimizations for querying changes, if you want to reference AI in AI, it seems like the way to go.
