Do Backlinks Still Matter? ChatGPT Now Favors Primary Sources

Link building eats a lot of time and money. Is it still worth it? Yes, but do less of it, do it better, and put the time you save into first-hand information that only you can provide. The latest data shows ChatGPT increasingly cites the original source of information, not the sites that repackage it.

Search Engine Journal reported Ahrefs' figures on October 12: from March to September 2026, the sites ChatGPT cited most in US answers shifted clearly toward primary, authoritative sources. The share of answers citing Reddit fell from 19.1% to 2.3%, and Wikipedia from 14.1% to 2.9%. PubMed climbed from 64th place to 4th, while WebMD fell from 18th to 79th.

Who does ChatGPT prefer to cite now?

By Ahrefs' count, the biggest climbers fell into three groups: government health and research sites, regulators, and reviewers who test products hands-on. The sites that lost ground were mostly health and finance sites offering secondhand interpretation; Britannica and Investopedia even dropped out of the top 200.

For sellers, "reviewers who test products themselves" making the list is a clear signal: real tests and real data are more likely to be cited than repackaged opinions. Keep the caveats in view, too. Ahrefs notes that Reddit and Wikipedia are still among the sites ChatGPT cites most, and that citation rank doesn't equal traffic.

The sample is large. Ahrefs analyzed the top 200 most-cited domains across about 2.4 million answers in March and about 3.8 million in September. To rule out noise from new questions, it also compared about 25,000 prompts that appeared in both months, and every site on the charts moved in the same direction.

Another number stands out: of 66.6 million free, logged-out ChatGPT responses, only 14.4% triggered a web search. In other words, a large share of answers rely on what the model already knows. Whether your brand information shows up consistently, over time, in trusted places will matter longer than how many links you pile up in one push.

Why does link building keep getting harder while paying off less?

That's the economic law of diminishing marginal returns: keep adding more of the same input, and each extra unit brings back less. Your first few links from industry publications and partners noticeably boost your site's credibility. By link number 200, you're likely paying for links on forums and directories that have nothing to do with your category, and your rankings barely move.

The time cost is easier to miss. Writing guest posts, mass-emailing for links and chasing replies can eat a whole week of one marketer's time. For a two- or three-person team, those hours could have produced a product test report or a clearer product page.

So real progress sometimes comes from stepping back. Stop doubling down where returns keep shrinking, and look at what AI and shoppers actually need: reliable, specific information that only you can give.

How should I split my time between backlinks and primary sources?

Here's our priority guide for small sellers, ranked by how much each move helps with AI citations and long-term brand credibility:

A simple way to rebalance: if you used to spend 70% of your SEO time chasing links, move half of that to the first two rows of the table. After three months, check how your AI citations and organic traffic have changed, then decide whether to keep adjusting.

How do I become a "primary source"?

1. Take stock of the information only you have. Factory test reports, material specs, quality-control processes, return-reason stats, frequent customer-service questions: many sellers have all of this, locked away in Chinese spreadsheets and internal chat groups.
2. Build an English test page. Spell out the method, the conditions and the results, for example "remaining battery after 8 hours of continuous use at -10°C," so the data can be verified.
3. Complete your official information. The spec sheets, certification numbers and warranty terms on your independent site should be the most complete and accurate version anywhere, and what you say on Amazon and social media should match it.
4. Publish small original research regularly. For example, send past buyers a short survey each quarter and turn the results into a report on "how US customers use this type of product." Content like this naturally gets cited and shared.
5. Trade data for links. Offer your test and research results to relevant media and reviewers. Good data earns links on its own; you don't need to buy them.
6. Recheck after model updates. Ahrefs suggests checking which sources ChatGPT cites on your core topics after OpenAI updates its models, then deciding what to fill in next.

A solid test page usually has five parts: the purpose of the test, the samples and environment, the test method, a results table, and conclusions with usage advice. Go easy on marketing adjectives like "ultra-strong" or "top-tier," and let numbers and conditions do the talking. Both AI and shoppers find that easier to trust.

Which backlinks are still worth building?

Real links with real context still have value: exhibitor pages from industry associations and trade shows, websites of partner brands and distributors, media coverage, and reviews by genuine testers. What they have in common is that they'd bring you potential customers even if SEO didn't exist.

What you can stop: bulk forum signatures, directories unrelated to your category, and link packages priced by the link. The test is simple. If you wouldn't bother reading the page a link sits on, US shoppers and AI probably won't take it seriously either.

What should Chinese sellers watch out for?

The most common problem is first-hand information that exists but can't be seen. A factory has done plenty of testing, but the documents exist only as Chinese PDFs. The English brand name is written one way on the independent site, another on Amazon and a third in trade-show materials. If AI and US shoppers can't see the information or match it up, all that data might as well not exist.

One more thing: many factory sellers treat certificates and test reports as "materials for procurement teams," sent out only when an inquiry comes in. To AI and to US end buyers, information that isn't public doesn't exist.

Turning that information into clearly structured English pages is usually the lowest-cost step you can take. If you want to know first who AI cites on your core questions, run a visibility audit with ToucHigh's GEO Autopilot, then decide where to start filling the gaps.