Reddit appears in 87.8% of US search results that show Google’s Discussions and forums module. Public Facebook Groups now appear in 38.3%, up from nothing at the end of 2024. Meanwhile the AI assistants that answer local questions are reading your reviews for tone, not just for stars. Three separate pieces of research published over the past fortnight point at the same conclusion: a growing share of what search engines and AI models believe about your business is written by other people, in places you do not own.
Where the forum data comes from
Glen Allsopp, Head of Marketing Strategy and Research at Ahrefs, worked with data scientist Xibeijia Guan on an analysis of more than 500 million search results. Matt G. Southern summarised the findings at Search Engine Journal. The numbers:
- The Discussions and forums module appears in 11.7% of US search results.
- Where it appears, Reddit is present in 87.8% of those results.
- Facebook is present in 38.3%, and every Facebook URL comes from a public Group.
- Quora is third globally at 32%, and every site below it sits under 10%.
- Facebook’s presence is spread widely rather than concentrated: its top 10 Groups account for 2.82% of its links, and its top 100 for 9.27%.
Facebook took second place from Quora in January 2026 and has held it since. It started from zero in late 2024, which is a fast climb for a surface most marketing teams do not track at all.
The practical recommendation in that research is the sensible one, and it is not “start a Facebook Group.” Allsopp suggests finding the Groups Google already surfaces for your queries and participating in those. Starting from zero on a surface where the top hundred Groups account for under a tenth of the links is a long road.
Is Reddit a shortcut to an AI citation?
No, and the person who wrote the most thorough guide to it this week opens by saying so. Marcella Merigo’s five-step framework at Search Engine Land starts here:
Reddit isn’t a shortcut to an AI citation. Posting promotional answers and waiting for Google or ChatGPT to notice them isn’t a community strategy. Reddit communities are quick to reject corporate messaging that contributes little of value.
Her framework asks you to define a “territory of authority” first: the intersection of what your audience needs help understanding, what your organisation knows from direct experience, and what your products or specialists can credibly speak to. Turned into a single question, that becomes a filter for which threads to enter and, more usefully, which ones to leave alone.
The line we keep coming back to is this one:
Search reveals intention. Community supplies context. Owned content provides depth. Monitoring shows whether authority is translating into visibility.
That is a clean description of why the channels stop being separable. A recurring question in a subreddit is evidence of a gap on your website. A repeated complaint is evidence that a product page is setting the wrong expectation. The language people use to describe a problem in a forum is frequently not the language your site uses, and the gap between the two is a content brief you did not have to commission.
Her advice on how to participate is worth stating because it runs against instinct: disclose the affiliation, answer the question directly, acknowledge limitations and trade-offs, link only when the resource genuinely adds something, and do not enter negative threads to correct every opinion. A response that admits the product is not right for every situation makes the rest of the answer more credible. That is uncomfortable and it is correct.
How do reviews feed what an AI says about a local business?
This is the part that has moved furthest, fastest. The Moz team’s piece on reviews and LLMs makes the mechanical argument: models do not simply compare average ratings, they read the text for sentiment and for detail.
The example given is a good one. Your website might list “HVAC repair” as a service. A review on your Google listing might say “they repaired my broken AC unit during the middle of a heatwave in under an hour.” That single sentence carries a service term, a product context, a sentiment and an outcome, and it does so in a customer’s own words. A model mining that listing for a query about emergency air conditioning repair has more to work with than your service page gave it.
Reviews are one of three places a machine checks your claims. The others are the directories and profiles you control, and the coverage you do not, which is the case for treating digital PR as an acquisition channel rather than a brand line. What we found about what actually earns links now applies directly: the same placements that earn a link are the ones an answer engine reads as corroboration.
Which leads to the obvious temptation, and to the reason it is now a mistake. In April 2026, Google added two clauses to the Maps user-generated content policy under the heading “Rating Manipulation.” The prohibited conduct now includes:
Merchants requesting that staff solicit a certain number of reviews.
Merchants requesting that staff solicit reviews that include specific content, including content that identifies a staff member.
Both of those describe programmes that were standard practice in local marketing eighteen months ago. Google still permits merchants to encourage customers to share a genuine experience, provided there is no incentive, no influence over the content, and no request for specific details to be included. The tightening also narrows what the review-removal industry can do, which we looked at when we tested one of those services. The distinction is between asking for a review and specifying the review.
The workable version, then, is to ask open questions rather than prescribe content. “How did we do?” or “Is there anything you would want another customer to know?” invites the detail without dictating it. Beyond that, most of what improves review text is not a review programme at all: it is doing the memorable thing that gives somebody something specific to write about.
One more finding from the same piece deserves attention, because it is where most businesses are still exposed: collecting reviews only on Google is a narrowing strategy. Yelp and OpenAI are working together, and the set of sources feeding local AI answers is widening rather than consolidating.
How much of this actually reaches AI answers?
Less than you would hope, which is exactly why the input quality matters. SOCi’s 2026 Local Visibility Index, cited by Search Engine Land, found that AI platforms recommend far fewer locations than Google’s three-pack does: 1.2% on ChatGPT, 7.4% on Perplexity, and 35.9% on Google.
The reason offered is profile accuracy, which averaged 68% on ChatGPT and Perplexity against 100% on Gemini, which pulls directly from Google Maps data. And the review threshold is visible in the same dataset: locations recommended by ChatGPT averaged 4.3 stars, those recommended by Perplexity 4.2. Review quality is functioning as a gate on recommendation, not merely as a ranking input.
A gate behaves differently from a ranking factor. A ranking factor lets you compensate elsewhere. A gate does not. The same dataset sits behind the wider decline in local clicks and calls.
What we would do first
Find out what is already there before adding anything. Search your priority topics, your brand and your comparison terms, and write down which Reddit threads, Groups, review pages and third-party discussions are already ranking. Then run the same set of questions through ChatGPT and one other assistant and record whether you appear, how you are described, and who is recommended instead. That baseline separates recognition from being understood, and the two are frequently confused. This is the same exercise as owning your brand search results, run against a wider set of surfaces.
Check your profile facts against your site and your schema. The 68% accuracy figure above is not a mystery. It is what happens when the hours, the services and the address disagree across sources.
Audit any review programme against the April 2026 policy. If a manager is asking staff to hit a review count, or to prompt customers to name them, that is now inside the prohibited scope. This is a quiet liability sitting in a lot of otherwise well-run businesses.
Pick two communities, not twelve. Depth in the places your buyers actually deliberate beats a thin presence across every subreddit adjacent to your category. Consistency is the thing being measured, by the community and by everything reading it.
Feed what you learn back into pages you own. The forum tells you the question. Your site is where the complete answer lives, and it is the only one of these surfaces where you control the wording.
Frequently asked questions
How often does Google show the Discussions and forums module?
Ahrefs’ analysis of more than 500 million search results found it in 11.7% of US results. Reddit appears in 87.8% of those, Facebook in 38.3%, Quora in 32% globally.
Do posts in private Facebook Groups show up in search?
The Ahrefs data found that every Facebook URL in the module came from a public Group. Posts from those Groups can appear in Google results for people who are not on Facebook.
Can I ask customers for reviews?
Yes. Google permits encouraging customers to share a genuine experience. What changed in April 2026 is that merchants may not ask staff to solicit a set number of reviews, and may not request that reviews include specific content such as naming a staff member.
Do star ratings decide whether an AI recommends a business?
Ratings are part of it, and text appears to matter as much. SOCi’s 2026 index found ChatGPT-recommended locations averaged 4.3 stars and Perplexity-recommended 4.2, while models also read review text for sentiment, services and outcomes.
The short version
Your reputation is increasingly assembled from sources you do not control and cannot edit. The response is not to try to control them. It is to be accurate everywhere a machine can check, to be genuinely present in the two or three communities where your buyers actually talk, and to earn reviews specific enough to be worth quoting.
