- Does this change really affect my company too, or just big publishers?
- What exactly happened: which sites gained and which lost?
- How does ChatGPT's citation mechanism really work?
- Three concrete scenarios: what changes for an e-commerce, a professional firm, and a manufacturing company?
- What content should I produce to be cited by ChatGPT?
- Why is ChatGPT's citation rate growing so fast?
- What are the risks and mistakes to avoid during this transition?
- How to get started concretely: the steps to adapt your content strategy?
- How to tell if my strategy is working?
- So, is it worth changing your content strategy now?
- How we see it at SHM
- Frequently Asked Questions
- ChatGPT is shifting citations from sites that summarize towards original sources: primary data, direct tests, institutional pages.
- Reddit went from 19.1% to 2.3% of cited responses in six months; Wikipedia from 14.1% to 2.9%. The model is changing fast.
- To stay visible on ChatGPT, you need content that only you can produce: proprietary research, product tests, internal data.
- 44% of citations come from the first third of the page: content structure and opening matter as much as the topic.
PubMed has jumped from 64th to 4th place among the sites most cited by ChatGPT in just six months. WebMD, which summarizes the same research, has slipped from 18th to 79th. If your company publishes content that explains what others have already said, this change directly affects you.
Does this change really affect my company too, or just big publishers?
It concerns anyone who publishes content online to be found: e-commerce, professional firms, manufacturing companies with a blog, agencies. If your texts explain topics that others have already covered better, ChatGPT will ignore them.
The change documented by Ahrefs and reported by Search Engine Journal it's not just about big media. It's about every company that uses content to get found online.
ChatGPT is stopping citing sites that summarize information already available elsewhere. Instead, it rewards those who produce original data, direct tests, and their own technical documentation. It's a shift in logic, not just in the algorithm.
To figure out if you're at risk, ask yourself this: does your content exist because you have something unique to say, or because you've rewritten what others have already published?
| Your situation | Visibility risk on ChatGPT | why |
|---|---|---|
| You publish general guides on common topics | No, for now | ChatGPT prefers the original source, not the summary |
| You have proprietary data, research, or internal tests | Yes | You are a primary source: ChatGPT tends to cite you |
| Manage an e-commerce site with detailed product pages | Yes | Brand pages are among the most searched with targeted queries |
| You have a blog with generic informative articles | No, for now | Without uniqueness, content is replaceable |
| Produce original case studies or industry reports | Yes | Content only you can create: direct advantage |
What exactly happened: which sites gained and which lost?
Between March and September 2026, institutional sites, regulators, and product testers rose in ChatGPT citations. Sites that explain and summarize dropped, some falling out of the top 200.
Ahrefs analyzed 2.4 million ChatGPT responses in March and 3.8 million in September, looking at the 200 most cited domains in US responses. The picture is clear.
PubMed (an archive of original medical research) has moved from 64th to 4th place. The FDA and other regulatory bodies have risen. RTINGS, which performs direct tests on electronic products, has gained positions. Britannica, Investopedia, and the New York Post have dropped out of the top 200.
Reddit went from 19.1% to 2.3% of cited responses. Wikipedia went from 14.1% to 2.9%. Both remain among the domains most retrieved by the model, but they are cited much less in the final responses.
Ahrefs verified that the shift is real: by comparing about 25,000 identical prompts present in both March and September, every site moved in the same direction. It is not an effect of changed questions, but of how ChatGPT chooses its sources.
| Site category | Direction | Examples |
|---|---|---|
| Original scientific research | ↑ Rising | PubMed |
| Regulatory and government bodies | ↑ Rising | FDA, CFPB, FTC |
| Direct product testers | ↑ Rising | RTINGS |
| Brand sites (apple.com, ikea.com) | ↑ Stable/rising | Official product pages |
| Sites that summarize medical research | ↓ Falling | WebMD, Verywell Health, Healthline |
| Sites that explain finance | ↓ Falling | Investopedia |
| Generalist encyclopedias | ↓ Falling | Britannica |
How does ChatGPT's citation mechanism really work?
ChatGPT doesn't cite everything it reads. It retrieves many pages in the background, but only brings a portion of them to the surface. Understanding this mechanism helps in creating content that gets past the filter.
When it receives a question, ChatGPT performs background searches before answering. These searches are called fan-out query : these are automatic queries that the model launches to gather updated information.
Out of 66.6 million free responses analyzed by Ahrefs between August and September, only 14.4% included a web search. When a search does occur, the model runs an average of 1.6 queries per response.
But retrieving a page doesn't mean citing it. According to an AirOps study on over 548,000 pages, ChatGPT cites only about 15% of the pages it retrieves : 85% is read but never appears in the final response.
An important piece of data for those who write content: 44.2% of citations come from the first third of the page , according to an analysis by Kevin Indig on 1.2 million responses. This pattern is called ski ramp : like a ski slope that starts steep and then flattens out, the value of a page for ChatGPT is concentrated at the beginning.
Put the most important and original information in the first few lines. Don't build the article like a funnel that leads to the conclusion: ChatGPT might never reach the end.
Three concrete scenarios: what changes for an e-commerce, a professional firm, and a manufacturing company?
The change has different effects depending on the type of content you produce. Here are three typical situations with priority actions.
Illustrative examples: the numbers and situations are meant to show the reasoning, they are not real customer data.
| Company type | Current content | Risk | Priority action |
|---|---|---|---|
| E-commerce (e.g., electronics) | Generic buying guides, copied product comparisons | High | Add internal tests, verified reviews, and your own return and satisfaction data |
| Professional firm (e.g. accountant) | Articles explaining regulations already published by the Revenue Agency | Medium-high | Directly cite the official source and add exclusive practical interpretation |
| Manufacturing company | Proprietary technical product sheets, certifications, laboratory data | Low | Structure product pages better and put technical data at the beginning |
For e-commerce, the risk is real: if your buying guides summarize technical specifications already available on the manufacturer's site, ChatGPT will cite the manufacturer, not you. The solution is to add data that only you have: real delivery times, return rates by category, and aggregated feedback from your customers.
For a professional practice, the path is different. You cannot compete with the Revenue Agency as the primary source of the law. However, you can be the primary source for practical interpretation: what that regulation means for an SME in your sector, with concrete numerical examples.
For a manufacturing company, the situation is often better than you might think. Technical data sheets, certificates of compliance, and internal test data are already primary sources. The problem is often the structure: this information is buried at the bottom of the page or in PDFs that the model cannot read.
What content should I produce to be cited by ChatGPT?
Ahrefs points in a clear direction: create pages that only you can create. Original research, direct tests, proprietary data, exclusive technical documentation.
ChatGPT's logic rewards those who are the source, not those who describe it. This doesn't mean you have to become a research institute. It means that every piece of content must have at least one element that doesn't exist anywhere else.
- 1 · Marketing / Content Identify what you know that others don't Internal data, conversion rates, customer feedback, test results: everything you have that isn't public elsewhere is a primary source.
- 2 · Marketing / Content Structure the content with data at the beginning 44% of citations come from the first third of the page. Put the most original and useful data in the first few lines, not at the end as a conclusion.
- 3 · SEO / Web Make data readable by the model HTML tables, structured text, correct semantic markup. PDFs and text images are not read by ChatGPT.
- 4 · Marketing / Content Cite the primary sources you use If you cite a study, link to the original source. ChatGPT tends to trace back to the primary source: help it find it through you.
- 5 · SEO / Analysis Monitor what ChatGPT cites for your topics Search for key questions in your industry on ChatGPT and see which sources it cites. That is the benchmark to beat.
| Content type | Value for ChatGPT | Practical example |
|---|---|---|
| Original research with your own data | Very high | Customer surveys, internal sales analysis by category |
| First-hand product testing | Very high | Direct product comparison with your own measurements |
| Exclusive technical documentation | High | Technical data sheets, certifications, user manuals |
| Practical interpretation of regulations | Medium-high | Operational guide on a regulation, with numerical examples |
| Article that summarizes other articles | Low | Roundup of others' opinions without original input |
| Copying information that is already public | Very low | Rewriting press releases or official regulations |
Why is ChatGPT's citation rate growing so fast?
In less than a year, the share of ChatGPT responses with at least one citation has increased more than fivefold. The phenomenon is accelerating and changing the rules of online visibility.
According to Similarweb data reported by Search Engine Journal , the share of US desktop ChatGPT responses with at least one citation went from 1.3% in June 2025 to 6.8% in May 2026. An increase of over five times in eleven months.
This means that more and more ChatGPT responses include links to external sources. Those who are cited receive direct visibility. Those who aren't, don't exist for those users.
The rate varies a lot by sector. Travel is the most cited, with 22.6% of responses including at least one link. Finance and Health are below the general average of 6.8%, despite being among the most searched sectors.
Another piece of data that changes the perspective: 65.3% of the pages most cited by ChatGPT come from domains with a Domain Rating of 81 or higher , according to an Ahrefs analysis of 1.4 million prompts. Domain Rating is a measure of a site's authority based on the links it receives from other sites. This doesn't mean that small sites are never cited, but that domain authority still counts for a lot.
To learn more about how visibility works on these tools, our cluster SEO and visibility for AI gathers all relevant updates for Italian SMEs.
What are the risks and mistakes to avoid during this transition?
The main risk is chasing ChatGPT citations with artificial or inflated content, losing credibility with real readers. Quality and uniqueness must go hand in hand.
Some common traps we see when companies try to adapt quickly to these changes.
| Common mistake | Why it doesn't work | What to do instead |
|---|---|---|
| Adding made-up or unverifiable data | ChatGPT tends to prefer verifiable sources; fake data harms your reputation | Use only real data, even if less impressive |
| Rewriting others' studies pretending you conducted them | The model traces back to the original source and cites it for you | Cite the source and add your own practical interpretation |
| Optimizing only for ChatGPT while ignoring human readability | Content designed only for AI models loses effectiveness with real readers | Write for people: AI models reward clarity |
| Wait for the model to stabilize before acting | Patterns change with every model update; those who wait lose ground | Monitor regularly and adjust incrementally |
| Focus only on ranking and not on citations | Being on the first page of Google does not guarantee citations on ChatGPT | Measure separately: search for your topics on ChatGPT and check who gets cited |
A point that is often underestimated: Ahrefs warns that its citation data does not measure traffic. Being cited by ChatGPT does not automatically equate to receiving visits. The value is in the visibility and perceived authority, not always in a direct click.
To understand how to concretely measure your visibility on AI, the article on how to get citations and measure visibility on AI Answers offers a practical method for SMEs.
How to get started concretely: the steps to adapt your content strategy?
Start with an audit of what you already have, identify what is unique, and restructure starting from the opening of the most important pages.
- 1 · Step 1 Audit existing content List the 10-20 most important pages for your business. For each one, ask yourself: is there something here that doesn't exist anywhere else? If the answer is no, that page is at risk.
- 2 · Step 2 Search for your topics on ChatGPT Type in the questions your customers usually ask. See which sources ChatGPT cites. Those sites are your direct competitors for AI visibility.
- 3 · Step 3 Identify the unique data you have Customer feedback, aggregated sales data, internal test results, direct experiences: everything you know that isn't public is an editorial asset.
- 4 · Step 4 Rewrite the openings of priority pages Put the most original and useful data in the first few lines. Don't build the article like a long introduction: ChatGPT mostly reads the first third.
- 5 · Step 5 Structure data in a readable format HTML tables, bulleted lists, structured text. Avoid putting key information only in PDFs or images.
- 6 · Step 6 Monitor and adjust every quarter Citation patterns change with every model update. Repeat the check on ChatGPT every three months for the most important topics.
An aspect worth keeping in mind: about 30 domains capture 67% of ChatGPT citations on a given topic , and the top 10 capture 46% of the total. The concentration is high. You don't have to be number one, but you need to be in the top thirty for your specific sector.
This is more achievable than it seems for an SME with a defined niche. An industrial valve manufacturer can be among the most cited sources by ChatGPT for technical questions about valves, even without competing with large generalist publishers.
The visibility strategy on ChatGPT is intertwined with the one for Google AI Overviews: the article on AI Overviews and brand visibility explains how the two logics overlap and where they differ.
How to tell if my strategy is working?
There is no single metric. You need to combine manual checks on ChatGPT, monitoring of AI referral traffic, and analysis of your ranking in responses for your key topics.
| Indicator | How to measure it | Frequency |
|---|---|---|
| Direct citations on ChatGPT | Manually search for key questions in your industry on ChatGPT and check if your site is cited | Monthly |
| Referral traffic from AI | Check your Google Analytics 4 for traffic coming from chatgpt.com, perplexity.ai, gemini.google.com | Monthly |
| Position in responses | Note if your site appears as the first, second, or third cited source for priority topics | Quarterly |
| Competitors in citations | Identify which sites are cited instead of yours for the most important questions | Quarterly |
| Content quality at the beginning | Verify that every priority page has original data in the first 200 words | At every content update |
Ahrefs reminds us that its ranking data does not measure traffic. Being cited is a signal of authority, but the commercial value depends on how many people see that citation and what they do next. Keep these two goals separate: AI visibility and actual traffic.
For those who want to understand how traditional SEO is transforming in this context, the article on SEO without Google index and site visibility for AIs offers a complementary perspective on how indexing is changing.
So, is it worth changing your content strategy now?
It is worth it if you already have content with original data, direct tests, or exclusive technical documentation. You just need to structure it better and put the most unique information at the beginning.
A middle ground You have a mix of original and summary content. Start with the topics most important to your business and add a unique element to every priority page, without rewriting everything.
Not yet if Your content is almost entirely summaries of information already available elsewhere. Before investing in new content, ask yourself what you really know that others don't: that's the foundation to start from.
How we see it at SHM
For us, this change confirms something we have long believed: technology rewards those who have something real to say. ChatGPT does nothing more than make visible a distinction that already existed: between those who produce knowledge and those who redistribute it.
Working with manufacturing and B2B companies on international positioning, we have seen time and again that highly specific technical content—detailed product sheets, test data, process documentation—performs better than any generalist guide, both on traditional search engines and, now, on AI. Not because they are better optimized, but because they are primary sources in the most literal sense: no one else can write them.
Our approach to SEO for AI it starts right here: identifying what a company knows that its competitors cannot copy, and building a content structure around that knowledge that AI models can read, understand, and cite. If you want to understand how to apply this to your case, the service of strategic AI integration is the most useful starting point. For those who also want to measure results over time, the service of Data processing and business intelligence allows you to track AI visibility systematically.
The change documented by Ahrefs is not an isolated novelty: it is part of a broader trend that we have also followed by analyzing how digital PR is becoming a tool to earn citations on ChatGPT and Perplexity . Those who build real authority today, reap the rewards tomorrow.
Frequently Asked Questions
Yes, but it's harder. 65.3% of the most cited pages come from domains with a Domain Rating of 81 or higher. However, for very specific and niche topics, a site with lower authority can be cited if it's the only primary source available on that subject. Vertical specialization partially compensates for lower general domain authority.
There is no fixed time: it depends on when ChatGPT updates its index and how often the model is updated. Ahrefs warns that patterns can change with every model update. A monthly manual check on ChatGPT for your key topics is the most direct way to see if the changes are having an effect.
No, the two logics are more similar than they seem. Both reward original, well-structured, and authoritative content. The main difference is in the structure: ChatGPT weighs the first third of the page heavily, while Google also evaluates overall coherence. Optimizing for AI also tends to improve traditional ranking.
Yes, but much less than before. Reddit went from 19.1% to 2.3% of cited responses between March and September 2026. Wikipedia from 14.1% to 2.9%. Both remain among the most retrieved domains in the background, but they are rarely cited in final responses. The citation rate for Reddit on retrieved pages is only 1.93%.
The easiest way is to manually search ChatGPT for the most important questions in your industry and check which sources are cited. You can also monitor referral traffic from chatgpt.com in Google Analytics 4: if you see sessions coming from that domain, it means someone clicked on a link that ChatGPT included in its response.
A primary source is whoever produces the original information, not whoever summarizes or explains it. PubMed publishes original scientific research; WebMD explains it. The FDA issues regulations; Investopedia explains them. For a company, a primary source means: internal data, tests conducted directly, exclusive technical documentation, proprietary research. Everything that cannot be copied because it exists only in your direct experience.
ChatGPT didn't change the rules of content: it made it impossible to ignore the ones that already existed.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
