ChatGPT's citation sources shifted hard in 2026, away from sites that summarize information and toward sites that produce it. Ahrefs Brand Radar data, as reported by Search Engine Journal, shows Reddit cited in 19.1% of US ChatGPT answers in March and 2.3% in September, while Wikipedia fell from 14.1% to 2.9%. The slots went to regulators, brand and support pages, and research databases, and the winners vary by category. The drop also belongs to one engine: Ahrefs' September table for Gemini still ranks reddit.com first, at 28.5% mention share.
If you spent the last two years building Reddit threads and Wikipedia-adjacent coverage, the headline is real. It also covers less ground than it looks. Before you re-plan, check what was measured, on which engine, and who now holds the slots in your category.
What the Ahrefs data actually measures
Ahrefs' October analysis, written by Louise Linehan, frames the change in one line: ChatGPT "seems to be moving from sites that summarize information to the sources that produce it." In March, about 1 in 5 US ChatGPT answers cited Reddit. By late September it was about 1 in 43. Wikipedia went from 1 in 7 answers to 1 in 35.
The scope matters as much as the size. These figures are the share of US ChatGPT answers citing a domain. They cover the 200 domains ChatGPT cited most often, from 2.4 million answers in March and 3.8 million in September. They say nothing about Gemini, Claude or Grok, about other countries, or about referral traffic.
You might suspect the prompt mix simply changed between March and September. Ahrefs checked: on roughly 25,000 prompts that were among its 150,000 highest-volume prompts in both months, every charted site moved in the same direction. The shift holds on identical prompts.
The winners show the pattern. PubMed climbed from 64th to 4th, and WebMD fell from 18th to 79th. Regulators such as the SEC, FTC and CFPB gained, as did support pages from Apple, Google and Microsoft and hands-on testers like RTINGS. Ahrefs' summary: "The pages holding up best are the ones only you can publish."
Why your tracker may show a different number
Here's where re-planning off the headline goes wrong. Ahrefs' own monthly ChatGPT table, labelled September 2026 and updated September 2, still shows reddit.com at #1 with 16.8% mention share. That isn't a contradiction. Mention share is a domain's slice of the top sources' citations, 2.3% is a share of answers, and the two very likely cover different time windows.
Trackers that measure share of all citations land somewhere else again. Promptwatch saw Reddit fall from 3.83% to 0.52% of ChatGPT Search citations. Brandlight measured 1.94% to 0.21% in a three-industry sub-panel, and Elmo recorded 4.67% to 0.38% in the ChatGPT UI. Every one points the same way, but none can be compared across tools. The fair comparison is a number against itself over time, on your own prompts.
It's a ChatGPT change, not an "AI" change
If the drop came from Reddit's side, every engine would show it. They don't. Otterly's analysis of 16 US brand reports found reddit.com in ChatGPT falling from 497 to 132 citations a day, down 73.4%. Across the other engines it tracks combined (Google, AI Mode, Gemini and Perplexity), Reddit went from 2,117 to 2,104 a day, a 0.6% dip. ChatGPT's total citation volume rose 3.5% in the same window, so it wasn't citing less overall. As Otterly puts it: "A change that shows up in one engine and not the others cannot be a change on the publisher's side."
The other trackers agree. Promptwatch saw only gradual declines on Google AI Overviews (about −11%) and AI Mode (about −30%), with no single break point. Brandlight, using a version-pinned control engine to measure normal drift, found ChatGPT's change was "nearly 20 times the drift floor." Ahrefs' September tables still put Reddit first on Gemini (28.5%) and Perplexity (21.6%), and its older June snapshot for Grok had Reddit first at 16.3%.
For a community marketer, that means your Reddit work may still be carrying three of the four engines your buyers use. We don't have a public 2026 number for Claude's Reddit or Wikipedia share, so treat Claude as unmeasured, not as safe. A blended "AI visibility" figure would have shown a mild dip and hidden a collapse on the most-used assistant.
The timeline, and what's still unknown
The change came in steps. On August 8, the share of ChatGPT fan-out queries (the sub-searches it runs behind a prompt) using the site: operator jumped from 0.37% to 16.8% in a day, and it has stayed around 16–17% since. Promptwatch's inference is that site-scoped searches typically target official sites, docs and government domains, and its co-founder Klaas Foppen reads it as "a change in how ChatGPT decides which sites to ask, not a Reddit-specific penalty."
That theory doesn't fully explain the bigger drop around August 14, and Promptwatch can't rule out a data-collection problem. OpenAI didn't respond to Gizmodo's request for comment. There's also a precedent. In September 2025, Reddit's ChatGPT share fell from about 7% to about 1%, then recovered to about 3% by late October (Profound data, via Elmo). Plan for this as the current state, not a permanent one.
Who took the slots depends on your category
So who moved into the citations Reddit and Wikipedia gave up? There's no single winner. Aleyda Solís, reading Similarweb data for August, put it this way: "Citations are no longer dominated by a few UGC sites and moving towards category specific sites."
The cleanest before-and-after comes from US personal finance. Green Flag Digital ran the same method on ChatGPT's top 500 cited pages in October 2025 and September 2026. Wikipedia and Reddit together went from 17.9% of citations to 0%. First-party brand pages rose from 1.8% to 22.4%, and regulators like the CFPB and FTC from 0% to 5.3%. Chase went from 4 cited pages to 63. The author's verdict: "a citation is rented, not owned."
Otterly's brand reports show the same pattern at company level: for Tesla, sec.gov and irs.gov rose and Reuters dropped out. Across its reports, official brand and competitor pages rose in 10 of 15, and news and media fell in 11 of 12.
For B2B, the nearest proxies are Brandlight's, pooled across all engines rather than ChatGPT alone. In business telephony, LinkedIn citations rose 194.1% while Reddit fell 35.3%. In shipping software, LinkedIn rose 83.3% and Reddit fell 37.3%. Without a per-engine split, this doesn't show LinkedIn gaining inside ChatGPT. It does show social citations in some B2B categories moving to another platform rather than vanishing. OpenAI's licensing partners gained too: their share of ChatGPT's 200 most-cited domains rose from 7.5% in March to 15.8% in July, and was still 12.5% in late September.
For B2B SaaS, the prompt type decides who wins
No public before-and-after study covers B2B SaaS specifically. The best guide is a separate Ahrefs test of 20,000 US prompts, which splits sharply by intent:
| Prompt type | Answers citing ≥1 official source | Share of citations to official sites |
|---|---|---|
| Factual (pricing, features, company details) | 84.4% | 63.4% |
| Comparison ("A vs B", "which is cheaper") | 68.3% | 46.4% |
On factual prompts, your pricing page, docs and help centre are now the default source, so a vague or gated pricing page hands the answer to someone else. On comparison prompts, most citations still go to sites other than the vendors being compared.
Ahrefs found this in its own category. For "AI visibility tool" prompts, review-type answers mostly cited reddit.com (56%), techradar.com (33%) and rankability.com (27%), and Ahrefs concluded it needs more reviews. The panel doesn't name the engine, but it shows Reddit still holding review intent in at least one B2B software category after the shift.
The pre-change baseline shows how much room vendors had. A DerivateX study published May 31 (40 categories, 10 runs each, ChatGPT with search) found that when ChatGPT recommended a tool, it cited that tool's own site only 11.6% of the time. In 25 of the 40 categories, no vendor was cited through its own site. As DerivateX put it, "Getting named is easy. Being the source is the real contest." The study is small and ChatGPT-only, but it suggests vendor sites had plenty of ground to gain on factual prompts.
Losing the link isn't losing the influence
There's a second reason not to scrap a Reddit programme: a citation records what got linked, not what shaped the answer. An earlier Ahrefs study of 1.4 million ChatGPT prompts, reported by SEJ, found ChatGPT using Reddit "extensively to understand topics, gauge consensus, and build context" while almost never crediting it. Pages from the dedicated Reddit source were cited only 1.93% of the time, yet made up 67.8% of pages retrieved but not cited.
Training data adds to this. Only 14.4% of the 66.6 million free, logged-out ChatGPT responses Ahrefs tracked ran a web search. The rest came from what the model already knows, and Promptwatch notes that Reddit threads may still shape training data even when no URL is cited. Paid and reasoning models search more, so buyers on paid plans may see a different mix.
So track mention rate and citation rate separately. Geoptimizer's own scan of notion.so shows how far apart they can sit: named in 92% of answers, cited in 3%. If your mention rate held through August while Reddit citations vanished, the community work is probably still paying off.
How to re-measure your citation slots, engine by engine
Only your own prompts tell you what happened in your category. Here's a re-measurement you can run in a week:
- Split your buyer-intent prompts by intent. Factual prompts (pricing, integrations, "does X do Y") go in one set. Comparison and review prompts ("best X for Y", "A vs B", "X alternatives") go in another. Blending them hides the signal.
- Log cited domains per engine. For each prompt, record which domains ChatGPT, Gemini, Claude and Grok cite, and whether your brand is named. Expect ChatGPT to look different from the other three.
- Compare like with like. Use your own pre-August history as the baseline if you have it, compare each metric only with itself, and use a rolling window. AI answers vary between runs, so four days of data is a snapshot, not a trend.
- Find who filled the gap. On ChatGPT prompts where Reddit or Wikipedia used to appear, note what's there now: a competitor's docs, a regulator, a review site, a partner publisher, LinkedIn. That list is your new target map.
- Act per intent. Where official pages now win factual prompts, make your pricing, docs and help-centre pages complete, current and readable by retrieval bots. Where third parties still win comparison prompts, keep up genuine Reddit and LinkedIn participation and earned reviews.
Two caveats about the measurement itself. First, the surface matters. Elmo found that before August, the ChatGPT UI showed about 11 times the OpenAI API's Reddit share, and after the change the gap closed. Geoptimizer queries the official APIs with web search on, and our methodology page says this matches the consumer apps "closely, but not exactly." Second, most of the data here comes from companies that sell tracking (Ahrefs, Otterly, Promptwatch, Brandlight, Elmo), and so do we. That's one more reason to check your own prompts.
That re-measurement is what Geoptimizer is built to run. Every prompt runs on all four engines, and each engine gets its own score with separate mention and citation rates. The trend chart marks the day an engine swapped models, so an August-style drop comes with a cause attached. The Starter and Pro plans re-check tracked prompts on a daily rotation.
FAQ
Did ChatGPT stop citing Reddit completely?
No. Reddit was still cited in 2.3% of US ChatGPT answers in September 2026, about 1 in 43, and Ahrefs says it remains among ChatGPT's most-cited domains. It's a steep cut, not a ban.
Why did ChatGPT's citation sources change in 2026?
The cause is unconfirmed. The leading theory is a retrieval change: on August 8, ChatGPT's site: fan-out queries jumped from 0.37% to 16.8%, pointing searches at specific domains. That doesn't explain the larger drop around August 14, and OpenAI hasn't commented.
Is Reddit still cited by Gemini and Perplexity?
Yes. Ahrefs' September 2026 tables rank reddit.com first on Gemini (28.5% mention share) and Perplexity (21.6%).
Should B2B SaaS companies stop investing in Reddit?
The data doesn't support that. Comparison and review prompts still cite third parties heavily, and Reddit shapes ChatGPT answers even when it isn't linked. Rebalance by prompt intent rather than pulling out. Our guide to earning Reddit AI citations without gaming it covers what genuine participation looks like.
What replaced Wikipedia in ChatGPT answers?
Primary sources in each category: brand and support pages, regulators, research databases like PubMed, and hands-on testers. Our Wikidata and entity data analysis cites Wikipedia at 8.9% of ChatGPT citations, an Ahrefs figure from July 2026 that predates this shift.
Re-plan what the data proves
The headline is accurate. On US ChatGPT answers, Reddit and Wikipedia gave up most of their citation slots between March and September 2026. But that's one engine, one country and one metric. Reddit still leads on Gemini and Perplexity, still holds review intent in B2B software, and what replaced it varies by category. Move factual answers onto pages only you can publish, the same lesson behind why original research earns AI citations, and keep your community work where comparison prompts still reward it.
Then measure it on your own prompts. Run your buyer prompts on all four engines with Geoptimizer, free with no credit card, and see who holds your citation slots today.