Your visibility in AI answers now depends on places you don't own. That feels uncomfortable, and it's exactly why so many good teams do something in a forum they later regret. This guide walks you through how to earn third-party mentions AI search engines actually pick up, without spam, sockpuppets, or a banned account. Nine steps, in the order you'd really do them.
You need less than you think to start: a few hours a week, one person willing to post as themselves, and patience.
Step 1: Draw the line between participation and manipulation
Before any tactic, get the vocabulary straight. Most teams cross the line by accident because nobody defined it.
Astroturfing is coordinated, sponsor-driven activity disguised as independent grassroots support. Undisclosed paid posts count. So do coordinated upvotes, AI-generated fake reviews, company-owned review sites that hide their ownership, and review pumping where an incentive is tied to a positive rating.
Sockpuppeting is running secondary accounts to upvote, agree with, or post about your own content. Reddit, Stack Exchange, Hacker News, Quora, and most professional communities all ban it outright.
An earned mention is a reference to your brand, product, or content that you neither paid for nor wrote yourself. That's the unit of value here. Everything else is a shortcut with a bill attached.
Four tests will settle almost every judgment call:
- Identity. A real name or one consistent handle is fine. Extra accounts are not.
- Intent. Contributing because the topic matters to you is fine. Only showing up when your brand can be named is not.
- Transparency. Disclose your employer or financial relationship. Some platforms require it. The rest reward it.
- Incentives. Anything of value traded for a review or mention has to be disclosed.
You'll know this step is done when someone on your team can look at a draft comment and say why it passes or fails, without asking you.
Common mistake: treating "we're not lying" as the standard. The standard is whether a stranger reading the thread would feel misled if they learned who you work for.
Step 2: Read each community's rules before you post
Every platform has already written down what it considers spam. Reading that page takes ten minutes and saves you an account.
Reddit publishes a self-promotion guideline usually summarized as 10:1: roughly one promotional submission for every nine or so genuine contributions. Individual subreddits stack more rules on top, and those are the ones that catch people. Minimum account age is often 7 to 30 days. Minimum karma is commonly somewhere between 50 and 500. Some ban specific domains outright or block links entirely for new members.
Quora collapses or removes answers that mostly promote a product without answering the question asked. Repeat behavior leads to downranking, removal, and eventually account restrictions. The community's rule of thumb lands near the same 9:1 spirit.
Hacker News allows Show HN posts, but expects you to describe how the thing was built or what problem it solved, not what it does for the buyer. Soliciting upvotes is a fast way to get flagged.
Stack Exchange is the strictest of the group. Self-promotion is expected to stay under roughly one in ten contributions, and you must disclose your affiliation every single time you mention your own product.
Gated Slack and Discord communities usually forbid solicitation altogether. Vendors are welcome when they show up as practitioners rather than sellers. Review platforms like G2, Capterra, Trustpilot, and Yelp prohibit paid or incentivized reviews and use behavioral signals (reviewer history, posting velocity, non-purchaser accounts) to flag what looks staged.
Here's the part worth remembering: you want to avoid spam AI citations never reward. Detection isn't a human reading your prose and judging your sincerity. It's pattern matching on same-device voting, content overlap between accounts, and coordinated timing. Sanctions escalate quietly, from a removed comment to a subreddit shadowban to a site-wide ban you may not even be told about.
You'll know this step is done when you have each target community's rules pasted into one shared doc, with the account-age and karma gates noted.
Common mistake: reading the platform's site-wide policy and skipping the individual subreddit or channel rules. That's where most bans start.
Step 3: Learn the legal floor, then disclose anyway
Platform rules are the ceiling most teams worry about. There's a legal floor underneath them, and in the US it got firmer.
The FTC's rule on fake reviews and testimonials (16 CFR Part 465) took effect on October 21, 2024. It prohibits buying or selling fake reviews, procuring them through officers or agents, review hijacking, insider reviews without disclosure, company-created review sites that hide the connection, reusing reviews across unrelated products, conditioning an incentive on a positive review, undisclosed incentives of any kind, and AI-generated fake reviews.
Civil penalties run up to roughly $53,088 per violation as of late 2025, adjusted for inflation. That figure is per violation, not per campaign. In December 2025 the FTC sent warning letters to ten businesses, so this is being enforced, not just published.
The rule doesn't require you to investigate every review that appears about you. It does expect action once you actually know something is wrong.
If you sell outside the US, similar regimes apply: the UK's CMA, the EU's Unfair Commercial Practices Directive, Australia's ACCC, and Canada's Competition Bureau.
Now the practical version. Ethical AEO off-site work has one habit at its center: say who you are, in the post, every time. Not in a bio nobody clicks. In the comment itself, in plain language. "I work at X, so take this with a grain of salt, but here's how we solved it."
That single sentence does more for your credibility than any tactic in this guide. Communities forgive a vendor who says so. They do not forgive one who gets caught.
You'll know this step is done when disclosure language is written, approved, and short enough that people actually use it.
Common mistake: assuming disclosure kills the mention. Disclosed contributions get quoted all the time. Undisclosed ones get deleted and screenshotted.
Step 4: Pick two or three communities, not twenty
You cannot be authentic in twenty places at once. Nobody can. The way you earn third-party mentions AI engines will actually surface is by being genuinely present somewhere, not thinly present everywhere.
Start by searching your highest-value buyer questions on Reddit, Quora, Stack Exchange, Hacker News, and at least one niche venue: a vertical forum, a Slack or Discord group, a Substack comment section, or GitHub Issues if you sell to developers. Sort what you find by real engagement, not size. Comment counts, accepted answers, thread age, and how often a thread gets updated tell you more than subscriber numbers.
Then check which engines matter for your category, because they don't agree. One analysis found that only about 11% of the domains ChatGPT cites are also cited by Perplexity. Perplexity cites Reddit in roughly 46.7% of analyzed responses, while ChatGPT leans hardest on Wikipedia, at around 47.9% in the same dataset. Coverage in one engine tells you very little about the others.
The good news for smaller brands: the citation pool has widened. Around 38% of Google AI Overview citations now come from pages already ranking in the top 10 organic results, down from 76% a year earlier. The other 62% come from somewhere else, and community threads are a large part of that somewhere.
This is where knowing which sources AI engines already pull from saves you months. DeepSmith's AI visibility module reports the sources AI cites most for your tracked prompts and shows which competitor pages win those citations, so you choose communities from evidence instead of instinct.
You'll know this step is done when you have a ranked shortlist of 10 to 20 places and have committed to just two or three.
Common mistake: defaulting to the biggest community. Smaller, tighter communities often produce the answers that get cited, because the answers are better.
Step 5: Earn standing before you ask for anything
This is the step everyone skips, and it's the one that decides everything else.
Spend several weeks contributing where your brand has nothing to gain. Answer questions in your area of expertise. Correct a detail. Share a template. Say nothing about what you sell.
Fill out your profile like a person, with a real bio and a handle you'll keep. If someone clicks your name, they should find a track record, not a launch announcement.
This is also the honest answer to how you get community mentions without getting flagged. Flagging is usually triggered by a thin account, not by a bad sentence. A three-month-old account with 40 helpful comments and one product mention reads completely differently from a three-day-old account with one link.
You'll know this step is done when you have a visible history of helpful, non-promotional contributions in each chosen community, and at least one person has replied to thank you.
Pro tip: track your own ratio before a moderator does. If your last ten contributions include two that name your company, you're already outside most platforms' guidance. Count in public terms, not intentions.
Step 6: Lead with value and let the mention come last
Now you can talk about your product. Carefully, and rarely.
Mention your brand only when it is the genuine answer to the question being asked. Frame it the way a builder would: "I built X because Y kept breaking," not "check out X." Add the disclosure. Then answer the rest of the question anyway, including the parts your product doesn't solve.
Keep the ratio above roughly nine helpful contributions for every promotional one, measured over time on each platform. That isn't a loophole to optimize. It's a floor.
Authentic brand mentions AI search engines can retrieve tend to share three traits: they're specific, they're attributed to a person with a visible history, and they sit in context next to a real problem. A vague "we love this tool" adds nothing an engine can extract. "We switched because our exports kept timing out at 50k rows, and here's what changed" is quotable.
Something else worth knowing: the most valuable mentions are usually the ones other people write about you. Your own disclosed comment is a data point. A stranger recommending you unprompted is evidence. Step 7 exists to make the second thing happen more often.
You'll know this step is done when at least one thread where you participated has a mention of your brand written by someone who doesn't work for you.
Common mistake: slipping a brand mention into every contribution. Detection systems catch the pattern regardless of how natural each individual sentence reads.
Step 7: Give people something worth mentioning
You can't earn a mention you haven't made easy. The fastest way to get named in a thread is to have published the thing people reach for when the question comes up.
Four asset types do most of the work:
- Original data. A study with a clean methodology and a table people can copy.
- A named framework. A process compressed into a memorable name and a simple diagram.
- Tools that finish a job. Templates, calculators, checklists, and one-click utilities that solve one specific problem.
- Expert commentary. A perspective or fact that isn't already sitting in twenty other posts.
These also give you a legitimate reason to pitch journalists through source platforms, with a short on-topic reply and two sentences of credentials rather than a product pitch.
Producing at this cadence is where most lean teams stall, because each asset needs research, structure, and a real point of view. DeepSmith's Content Studio turns a planned idea into a finished, brand-grounded article with citation-ready structure built in during writing, so the publishing side of this loop keeps moving while you're busy being useful in communities.
You'll know this step is done when at least one asset per quarter gets referenced, linked, or quoted by someone outside your company within 90 days.
Common mistake: publishing another opinion post and hoping. Opinions rarely get cited. Numbers, frameworks, and tools do.
Step 8: Write your guardrails down as a team policy
Everything above lives or dies on the day someone improvises. A written policy is how you make ethical AEO off-site behavior survive a bad quarter.
Keep it to one page. Cover six things:
- Who posts. Named people, under their real identity, no shared or anonymous accounts.
- Disclosure. The exact sentence, and the rule that it appears in the post, not just a profile.
- Incentives. No gift, discount, or sweepstakes entry tied to a review or mention, ever, disclosed or not.
- Voting. No coordinated upvoting, including asking customers in a Slack group. Well-meant coordination still triggers site-wide bans.
- Agencies and contractors. The same rules apply to anyone you pay. The FTC rule reaches activity procured through agents.
- Escalation. Who gets called when a moderator pushes back, and who approves the response.
Take the reputational risk seriously, because it is asymmetric. Astroturfing campaigns get exposed in public: threads, screenshots, press coverage, and in some cases regulator settlements with US states over undisclosed grassroots messaging. The upside of a manipulated mention is one citation. The downside is a permanent search result about your company's integrity, and AI answers that now repeat it.
Every rule on this list exists to help you avoid spam AI citations quietly punish anyway. You're not giving up leverage. You're removing the failure mode that erases the leverage you build.
You'll know this step is done when the policy is signed off, shared with agencies, and someone owns it by name.
Common mistake: writing the policy and never testing it. Walk a real borderline example through it once, out loud, with the team.
Step 9: Track which mentions turn into AI citations
Off-domain work compounds slowly, which makes it easy to abandon right before it starts working. Measurement is what keeps you honest and funded.
Track five numbers monthly:
- Mention rate. The share of your tracked prompts where AI names your brand.
- Citation rate. The share where a page you own is cited as a source.
- Share of voice. Your mentions against the total across your tracked competitor set.
- Earned mention volume. Third-party mentions per month across your chosen communities.
- Sentiment. The positive, neutral, and negative split of those mentions.
Notice what's missing: sessions. About 20% of ChatGPT brand mentions include a clickable citation link visible in analytics, which means roughly 80% leave no trackable click at all. If you judge this work by referral traffic, you will conclude it failed while it's actually working.
The engines also move. Reddit's share of AI citations grew roughly 73% between October 2025 and January 2026 across tracked categories, with some verticals more than doubling. A source mix that's true today may not be true in two quarters, so re-baseline rather than assume.
You can start free: run a fixed set of 20 to 30 prompts through each engine once a month and log the results in a shared sheet. When that becomes the bottleneck, DeepSmith tracks mention rate, citation rate, and share of voice across ChatGPT, Perplexity, Gemini, Claude, and Google AI Mode with full answer history, and coverage scales with your plan.
You'll know this step is done when you can show a monthly trend, not a snapshot, across at least two consecutive quarters.
Common mistake: changing the prompt set every month. Freeze it, or you're measuring your own edits.
What to do next
Pick one community this week. Just one. Read its rules, fill out your profile properly, and answer three questions without mentioning your company at all.
That's the whole first move. It feels too small to matter, and it's the reason the teams who do this well have accounts nobody can dispute.
Do it again next week. In a quarter you'll have something no shortcut produces: a real reputation in a place AI engines actually read.
If the publishing side is what's holding you back, DeepSmith runs tracking and production in one platform, so you can see which prompts you're missing and produce the assets that earn authentic brand mentions AI search engines can retrieve. Start a free trial and see your real numbers before you commit.
