In this business, we have always known that a spontaneous recommendation in a forum could be worth more than ten paid links. We knew it before there was an acronym for it, back when forum marketing was a craft built on patience, context, and a discreet signature.
The principle has not changed. What changed is who is reading those conversations.
Today, AI-powered search engines read them too. And Reddit occupies an especially visible place: a Semrush study found that it appeared in 12.6% of the ChatGPT Search responses analyzed, 9% of Google AI Mode responses, and around 4% of Perplexity responses.
When someone asks an AI which tool to use, which hotel to book, or which company to trust, there is a real possibility that part of the answer will be based on a Reddit thread.
The strategic conclusion was immediate: you need to be part of those conversations.
The problem is that Reddit came to its own conclusion too.
Why Forum Mentions Carry So Much Weight in GEO
When we audited ChatGPT Query Fan-out, we saw that a single query can expand into several secondary searches. Some verify facts; others look for comparisons, experiences, objections, and recommendations.
This is where forums gain ground.
A corporate article says what a company wants to say about itself. A thread with questions, answers, and disagreements shows how different people interpret a product or category.
For a system that needs to build a recommendation, that conversation can function as a layer of human consensus.
That does not mean models consider everything published on Reddit to be true. It means they find signals there that are difficult to reproduce on a commercial page: concrete experiences, comparisons, spontaneous language, and real objections.
The Semrush study provides another interesting data point: more than half of the citations coming from Reddit corresponded to question-and-answer threads. In addition, most of the cited posts were not viral. Eight out of ten had fewer than twenty upvotes, and seven out of ten had fewer than twenty comments.
The ability to be cited does not seem to depend only on popularity. The clarity of the question, topical relevance, and usefulness of the answers also matter.
The environment surrounding the mention matters just as much as the mention itself. We already analyzed this with anchor context: the signal does not live only in the link or brand name, but also in the words and entities around it.
A recommendation in a forum does not only say which company to choose. It also explains what it is useful for, which alternatives it competes against, what problem it solves, and what objections appear around it.
So much for the theory. Now, the trenches.
The Problem: Participating on Reddit Has Become an Obstacle Course
Anyone who has tried to work on mentions on Reddit knows a similar sequence.
You write a reasonable comment, provide useful information, and avoid including links. Shortly afterward, it stops receiving impressions. In some cases it was removed. In others, it remains visible from the account that published it, but not to everyone else.
Shadow removal is an especially efficient way to waste time.
There are several overlapping layers of control:
- Automated rules for each subreddit: minimum karma, account age, blocked domains, restricted words, and filters that are not always public.
- Human moderation: an account that only participates when a commercial opportunity appears is easy to recognize.
- Reddit’s automated controls: the platform is expanding its systems to detect spam, artificial activity, and coordinated patterns.
In July 2026, Reddit explained that it analyzes signals from the moment an account is created and that it uses language models to detect fake behavior that its previous systems could not identify.
It also reported that it was detecting around 25,000 new spam posts and comments per day, blocking millions of views of unwanted content, and invalidating inauthentic votes.
Reddit did not present these measures specifically as an offensive against GEO. Its stated objective is to reduce spam and manipulated activity. But it is reasonable to infer that operations designed to manufacture recommendations can also be affected.
The dividing line is no longer whether or not a link is included. It is the behavior of the account.
A conversation can mention a brand and be legitimate. It can also contain no links at all and still be part of a coordinated operation.
The practical result is clear: landing in someone else’s threads with pre-written comments has an increasingly short life expectancy. The cost of insisting is no longer just losing a post. It can also mean losing the account and all the history built around it.
The Change in Approach: Stop Chasing Mentions and Start Creating Conversations
Faced with this situation, our team started testing a different approach: instead of looking for opportunities to insert a brand into other people’s conversations, identify genuine questions that still do not have a satisfactory discussion and open a new thread.
The difference may seem small, but it changes the nature of the action.
The objective stops being getting a mention and becomes creating a conversation that deserves to exist.
This approach offers some advantages.
The Initial Post Defines the Problem
The person opening the thread can raise a specific question, explain a real case, and organize the context from the title and initial body of the post.
Question, comparison, and discussion formats are precisely the ones that appear most frequently among Reddit posts cited by AI-powered search engines.
The Semantic Context Is Deeper
In an isolated comment, the mention depends on the framing defined by someone else. In your own thread, the central problem can be developed from the beginning.
That does not mean controlling what the community will say. In fact, the conversation may contradict the initial hypothesis, criticize a company, or recommend alternatives. That lack of control is also part of what makes it credible.
The Thread Can Become a Citable Unit
A well-defined thread brings the question, experiences, comparisons, and objections together under a single URL.
There are no guarantees of indexing or citation. But there is a more coherent thematic unit than a standalone comment inside a generic conversation.
What We Are Observing
It is worth being clear about the current state of this test.
We have been running it for a relatively short time and on a limited number of cases. So far, threads centered on real questions survive and generate more participation than comments inserted into other people’s conversations.
This is a field observation, not a statistical study.
We still need to measure how long they remain visible, how many are indexed, and whether they actually appear as sources in ChatGPT Search, Perplexity, or Google AI Mode.
The hypothesis is promising, but it is still a hypothesis.
Conditions for This Approach to Work
Some conditions seem non-negotiable.
An Account With Genuine Participation
An account created to post about a single company reproduces exactly the pattern anti-spam systems are trying to detect.
Account age helps, but it is not enough. What matters is accumulated behavior: normal participation, diverse interests, and contributions that do not always pursue the same objective.
This should not be confused with “warming up accounts.” Manufacturing activity to simulate a genuine identity is still manipulation.
A Question That Stands on Its Own
An ad disguised as a question is still an ad.
The question should be useful even if the brand never appears. It has to allow for different answers, criticism, and alternatives.
When the thread only leads toward one company, the intention becomes too obvious.
The Brand Appears Sparingly and When Relevant
Not every answer needs to mention a product.
The company should appear when it adds information, solves a specific part of the problem, or helps compare alternatives. When every contribution returns to the same name, the thread stops looking like a conversation.
Transparency When There Is a Commercial Relationship
When the person participating works for the company, represents a client, or has a financial interest, hiding it can turn a valid contribution into a deceptive practice.
A transparent mention may be less persuasive in the short term. It also has a better chance of surviving when someone reviews the account’s full history.
Accept That the Conversation Cannot Be Controlled
Opening the thread does not make the author the owner of the answers.
Managing it means answering questions, acknowledging objections, and accepting that some users will recommend competitors.
Abandoning the conversation as soon as the desired mention appears makes the intention far too obvious.
Is Creating Your Own Threads Really Safer?
Not necessarily.
Your own thread can be removed just as quickly as a comment if it repeats an existing question, uses a suspicious account, or tries to artificially steer the discussion.
The advantage is not in being the OP. It is in changing the objective.
When the strategy is still to manufacture a recommendation, opening the thread only moves the manipulation to another part of the page. When the strategy is to document a real question and gather experiences, the content has a reason to remain published.
That difference also matters for GEO.
AI-powered search engines do not only need brand names. They need sources that allow them to build useful answers.
A post that survives is not automatically one that will be cited. And a post that is cited does not guarantee that the brand will be recommended in the expected way.
Conclusion: The Mention That Survives Is the One That Earned the Right to Exist
There is something circular about the current moment.
Community platforms are valuable to AI-powered search engines because they contain human conversations that are difficult to find elsewhere. For that same reason, they need to defend themselves against those trying to manufacture those conversations at scale.
For those of us who have worked in search for years, the lesson feels familiar. Every time a valuable channel appears, the industry tries to industrialize it. Then come saturation, filters, and penalties.
It happened with directories, guest blogs, and author profiles. Now it is happening with forums.
In this scenario, creating your own threads can work. But not because the OP has any special immunity or because it allows you to control the context.
It can work when the question is legitimate, participation is transparent, and the conversation still has value even if the brand disappears from it.
The best mention for GEO is not the one that managed to fool the moderator.
It is the one nobody has a reason to delete.
Sources
- Reddit Inc. — How We’re Keeping Reddit Real and Safe in the AI Era, July 2026.
- Semrush — We Analyzed 248K Reddit Posts: What Drives Visibility in AI Search.
- Reddit Inc. — Reddit Reports Fourth Quarter and Full Year 2025 Results.
With a solid professional track record, Luciano has been SEO Manager at MD Marketing Digital since May 2018, where he leads a talented team, focusing on search engine optimization and the implementation of effective digital strategies.
Leave a Reply