How to rank on ChatGPT now that it has stopped citing Reddit
Most advice on getting into ChatGPT’s answers says to get mentioned on Reddit. In August, ChatGPT largely stopped citing Reddit. We ran our own test to see what it relies on now.
For two years the standard advice for showing up in ChatGPT’s answers was simple: get people talking about you on Reddit, because ChatGPT cites Reddit constantly. That advice is now out of date.
What changed in August
In the second week of August 2026, ChatGPT’s citations of Reddit fell sharply. OtterlyAI, tracking 16 US brands, measured reddit.com citations dropping from 497 to 132 per day, at least 73%, in two steps on 8 and 14 August. ChatGPT’s total citations rose slightly over the same period, so it was citing as much as before, just different sources. Qwairy’s analysis, comparing the same brands and prompts before and after, put the drop at 95%; the numbers differ by method, the direction doesn’t.
Three details matter for anyone doing GEO (generative engine optimisation):
- It was ChatGPT only. In OtterlyAI’s data, Google’s AI features, Gemini and Perplexity kept citing Reddit at almost exactly the same rate. Qwairy, with a mostly European sample, also found Google AI Overviews flat, though it saw Google AI Mode fall by about 44%.
- Official and reference pages gained. Brand websites, encyclopedias and other commercial sites took much of Reddit’s share.
- The recommendations didn’t change much. Mi3 quoted Petra Labs co-founder Sami Akkawi:
We did not notice meaningful differences in the brands being recommended.
Users see the same brands; the link in the citation now points somewhere other than Reddit.
OpenAI has not explained the change. The rest of this post is what we saw when we tested it ourselves.
Our test: what ChatGPT cites now
On 10 October 2026 we asked ChatGPT two buying questions in a category we have studied closely, meal kits. The answers came as product cards with prices and a short list of trade-offs, each with small source links.
| Question | Recommended | Sources shown |
|---|---|---|
| Best meal kit for one person (US) | EveryPlate, then Dinnerly, HelloFresh, Blue Apron | Mostly each brand’s own website, plus a third-party site for one |
| Best low FODMAP meal delivery | ModifyHealth, then Epicured | ModifyHealth’s website, Monash University’s FODMAP programme, NBC Select |
No Reddit links in either answer. Two other things stood out.
The first answer was for the wrong country. Asked in English from outside the US, ChatGPT recommended Japanese services with prices in yen, then asked us to confirm budget, cooking time and delivery country before answering for the US. AI answers are personalised by location and by follow-up choices. One screenshot of one answer is not a ranking, for you or for a competitor.
The answer was organised around buying criteria, not brands. For the one-person question, ChatGPT worked through cost per serving including shipping, minimum portions, cooking time and delivery coverage, and warned that the advertised starting price is not necessarily what you pay. For the FODMAP question it separated the introductory offer from the regular price and pointed to third-party certification.
Those criteria came from real customers first
We had read those exact concerns before, in a 325-comment meal kit thread we used in How to find content ideas. Set side by side:
| What customers asked each other | What ChatGPT’s answer covered |
|---|---|
How much per meal after the discounts run out?(link) | Regular price versus introductory offer; total cost per serving after shipping |
which kits are the most single-friendly?(link) | Most kits sell two-serving recipes; cook two and keep one for lunch |
I can't do onions or garlic(link) | Low-FODMAP services that let you exclude ingredients |
And the brand ChatGPT picked for low FODMAP is the one a customer recommended in that thread 21 months earlier: I do recommend ModifyHealth! I can't eat onions or garlic, so I order their low FODMAP meals.
(link)
We can’t see inside ChatGPT, and one example proves nothing about cause. But it fits the pattern described above: the sources ChatGPT links to changed, and the brands it recommends did not.
What to do now
If ChatGPT matches products to buying criteria and checks the facts on official and third-party pages, the work splits into two parts: know the criteria, and make sure the facts are easy to find.
1. Find the criteria your buyers actually use
Read the discussions where people choose between products in your category and list what decides it for them, in their words. Count how many different people raise each one. For meal kits it was real cost, portions for one, cooking time, dietary limits and cancelling. For your category it will be different, and it is rarely the list on your homepage. We showed the method in Voice of customer analysis: a worked example.
2. Answer each criterion plainly on your own site
Official pages gained the most from August. Make sure a reader (or a model) can find, without signing up: the regular price after any introductory offer, what is included, minimums, who it suits and who it doesn’t, how to cancel, and where you deliver or operate. In our test, ChatGPT had to qualify one brand’s headline price because the starting price was not necessarily what a customer would pay. Vague pricing becomes a caveat in the answer.
3. Earn third-party proof
The low-FODMAP answer leaned on an independent certification body and a media review, not just the brand’s claims. Certifications, standards bodies, and reviews by publications that test products give the answer something to check your claims against.
4. Don’t try to plant Reddit mentions
Seeding Reddit with brand mentions was always risky: communities remove it, and it damages trust when spotted. Now it also does less for ChatGPT specifically. Public discussion still matters, and other AI tools still cite it, but as something to learn from, not something to manufacture.
5. Measure answers, not rankings
Ask the questions your buyers ask, regularly, in more than one assistant, and from the right country. Record what is recommended and what is cited. Expect variation; look for patterns over weeks, not a single answer.
About this test. We asked ChatGPT two questions on 10 October 2026; answers vary by location, account and follow-up, and prices shown in the answers are ChatGPT’s, not verified by us. Citation statistics are from the third-party analyses linked above, which use their own samples and methods. Reddit quotes are verbatim and link to their source; we don’t show usernames. Valtaris is not affiliated with OpenAI, Reddit or any company mentioned.