Signals
The 6 types of AI citation your brand needs, and how to earn each one
By Arnav Mukherjee, founder of TofuBofu · August 5, 2026
Almost every post we have written on this site, including several of mine, ends at the same instruction: build corroboration. Get independent sources saying consistent things about you. It is correct, and on its own it is close to useless, because nobody has told you what the sources actually are or which ones are worth your Tuesday.
So this is the missing list. Six types of citation, in the order I would earn them, with what each one is, why an engine weights it, how to actually get it, and roughly how long it takes to show up.
One framing note before the list, because it changes what you do with all six. These are not backlinks. A backlink passes authority through a hyperlink. A citation passes corroboration through a statement, and it does not need to be linked at all. An unlinked mention in a Reddit thread can do more for an AI answer than a linked profile on a page no human reads.
The six types, by effort and durability
1. Your own structured pages (the floor, not the win)
What it is. The pages on your own domain that state plainly what you do, for whom, where, and at roughly what cost, marked up with Organization, Service and FAQPage schema so a machine can parse the claims.
Why engines weight it. They need something to retrieve before they can repeat anything. SE Ranking found structured data on 71 percent of the pages ChatGPT cites, and 65 percent for Google AI Mode. This is also the only citation type you fully control.
How to earn it. Write the specifics you have never written down: cities served, industries you are genuinely good at, client size range, compliance posture, response commitments, pricing model. Then mark them up. Prose first, schema second, because schema on a page that says nothing has nothing true to carry.
Timeline, and the catch. Days to a few weeks once crawled. The catch is that this alone does not work. We enriched 72 IT firms already measured as missing from AI answers, and 28 of them ran their site on HubSpot CMS with total control of the crawlable surface. They were still absent. Your own pages say what you claim. The next five say whether anyone agrees.
2. Directory listings (necessary, and badly oversold)
What it is. Clutch, G2, Capterra, your industry and regional association directories, your Google Business Profile.
Why engines weight it. Directories are structured, current and independent, which makes them cheap corroboration that you exist and do what you say. Profound found Clutch accounts for roughly 2.4 percent of ChatGPT citations, which is small but real for services firms.
How to earn it. Claim the profile, then make the fields match your site exactly. Same legal name, same city, same category. Inconsistency across profiles is worse than absence, because it makes you harder to resolve as a single entity.
Timeline. Two to six weeks.
Do not overrate this one. In our fifteen-region study we checked every company the engines named against that region's own Clutch directory. 150 of the 228 directory-listed firms were named by no engine at all. Being listed did not get them recommended. Treat directories as the second source that confirms you, never as the thing that earns the mention. Full working in AI keeps recommending the big names.
3. Review platforms (specifics beat volume)
What it is. G2, Capterra, Trustpilot, Google reviews, and the review sections inside directories.
Why engines weight it. Reviews are independent text about you at volume, which is exactly what a model needs to justify a recommendation. G2 alone holds about 22.4 percent share of voice for software, per Kevin Indig's analysis with G2.
How to earn it. Ask at the moment of delivered value, not at renewal. Then ask for specifics rather than praise: which industry they are in, how big their team is, what problem you solved. A review that says great team is worth almost nothing to an engine. A review naming your vertical and the outcome is quotable.
Timeline. Three to eight weeks, and gentler than the platforms imply: roughly 10 percent more reviews is associated with about 2 percent more AI citations.
The line we will not cross. Fabricated reviews are off the table. They violate every platform's terms, they are illegal under FTC rules, and they get wiped when detected, taking your profile with them. We were asked about this once and declined, and the reasoning has not changed.
4. Community threads (the highest-variance type)
What it is. Reddit, Hacker News, industry Slack and Discord archives, Quora, niche forums. Somebody asking for a recommendation and somebody else naming you.
Why engines weight it. Heavily, and more than most B2B marketers expect. Profound found Reddit accounts for roughly 12 percent of ChatGPT citations. Google also signed a content licensing deal with Reddit, so this is structural rather than a passing quirk. The value is that it is unprompted opinion, which is the closest thing to a real reference an engine can find.
How to earn it. Be useful in the threads where your buyers already are, under your real name, answering questions without pitching. The version that works is a founder giving away the answer. The version that gets you banned is a new account recommending your own product, and moderators are extremely good at spotting it.
Timeline. Unpredictable. A single thread can surface in AI answers within weeks or never. This is the type to run continuously in the background rather than as a campaign with a deadline.
See which sources AI is already using about you
A free scan across six engines shows the citations behind each answer, so you can see which of these six types you are missing.
Get your free audit5. Comparison pages (yours and other people's)
What it is. Pages that put you next to a named alternative. Your own vs-competitor and alternatives-to pages, plus any third-party page that compares you.
Why engines weight it. Because they match the structure of a comparison question exactly. When a buyer asks how you differ from a competitor, a page already answering that is the most directly usable source in the index. These pages also attach your name to a competitor's entity, which is how you get pulled into answers about them.
How to earn it. Write them yourself, honestly, and include the cases where the competitor is the better buy. Engines quote specifics, so a page that concedes a real weakness reads as more credible and gets quoted more than a page claiming you win on everything. Keep the pricing current and date it, because a stale comparison is worse than none.
Timeline. Four to twelve weeks, and these compound. We maintain a set of these and the honest ones outperform: see our Otterly alternatives page, where the competitor pricing is dated and read from each vendor's own site.
6. Third-party roundups (the hardest, and worth the most)
What it is. Somebody else's best-X-for-Y listicle with you in it. An industry publication, a niche blog, an analyst roundup, a community wiki.
Why engines weight it. More than any other single type, because the page is already shaped like the answer. A buyer asks for the best vendor in a category and a ranked list of vendors in that category is the ideal retrieval target. One placement can put you into answers for dozens of related questions.
How to earn it. Find the roundups that already rank for your category question, then give the author a reason to add you: a specific, checkable claim, a real client outcome, a data point they cannot get elsewhere. Original data is the strongest currency here, which is most of why we publish our own research. Paid placement exists in this category and often arrives disguised as editorial, and I would avoid it. An undisclosed advertorial is a disclosure problem in most jurisdictions, and readers work it out.
Timeline. Two to six months to land the first few. The longest-lived of the six once you are in.
The order I would actually work them
Breadth beats depth, and this is the one number worth memorising: Profound found brands appearing across four or more platforms were about 2.8 times more likely to be cited. Four thin presences beat one immaculate one. So the sequence is about getting to four types quickly, not perfecting the first.
Weeks 1 to 2: your own pages
Write the specifics and mark them up. It is the only type you fully control and everything else assumes it exists. Do not stop here, which is the mistake most firms make.
Weeks 2 to 4: directories and review profiles
Claim, correct, and make every field match your site exactly. Then start asking for reviews at the moment of delivered value, asking for specifics rather than stars.
Ongoing from week 1: community presence
Start immediately because it has the longest and least predictable lag. Half an hour a week in two subreddits where your buyers actually post, answering rather than pitching.
Weeks 4 to 8: your comparison pages
One page per named competitor, honest about where they win, with dated pricing. These are also the pages that convert, so the work pays twice.
Month 2 onward: earn the roundups
Build one piece of original data worth citing, then approach the roundups that already rank for your category question. This is slow, and it is the one that compounds.
Measure it per engine, not on an average. Our own research across five metros found 85 percent of firms were named by exactly one engine out of four, so a blended score will tell you almost nothing about whether any of this worked. Re-run the same buying question, hold the wording fixed, and watch which engines start naming you.
Frequently asked questions
What is an AI citation?
An AI citation is any independent source an engine draws on when it describes or recommends your company. It is broader than a backlink, because it does not need to be a link at all. A Reddit comment naming you, a Clutch profile, a G2 review, a roundup that lists you among five vendors, or a competitor's comparison page that mentions you all qualify. What they have in common is that somebody other than you said something specific about your company in a place a model can retrieve.
Are backlinks the same as AI citations?
No, and treating them as the same is the most common mistake we see. A backlink passes ranking authority through a hyperlink. An AI citation passes corroboration through a statement, whether or not it is linked. This is why an unlinked brand mention in a relevant Reddit thread can do more for an AI answer than a linked directory listing on a page nobody reads. Backlinks still matter for the search rank underneath, so the honest position is that they are demoted rather than dead.
Which type of AI citation is worth the most?
Third-party roundups, meaning the best-X-for-Y listicles, tend to be worth the most per placement, because they match the exact shape of the question a buyer asks an engine. When someone asks for the best vendor in a category, a page that already answers that question with a ranked list is the most directly usable source available. Community threads come second because they are dense with unprompted opinion. But no single type carries an answer on its own, and breadth beats depth: Profound found brands appearing across four or more platforms were about 2.8 times more likely to be cited.
Do directory listings still work for AI visibility?
They are necessary and badly oversold. Our own regional research makes the point bluntly: across fifteen regions we checked every company the engines named against that region's own Clutch directory, and 150 of the 228 directory-listed firms were named by no engine at all. Being listed clearly did not get them recommended. Directories work as corroboration, meaning the second or third source that confirms you exist and do what you claim, rather than as the thing that earns the mention.
How many reviews do I need for AI to notice?
There is no threshold, and the relationship is gentler than review platforms imply. Kevin Indig's analysis with G2 found roughly 10 percent more reviews associated with about 2 percent more AI citations, so it is a real but incremental effect. What matters more than count is that the reviews contain specifics an engine can quote, such as your industry, company size served, and the problem you solved. Ten detailed reviews naming your vertical beat a hundred that say great team.
How long does each type of citation take to work?
Roughly, and from our own scan cadence rather than a vendor promise: your own structured pages can be picked up within days to a few weeks once crawled, directory and review profiles take a few weeks to surface, community threads are unpredictable and can appear fast or never, and third-party roundups are the slowest to earn but the longest-lived once you are in one. Set the expectation at one to three months before a re-scan shows movement, and measure per engine rather than on a blended average.
Can I just publish more content on my own site instead?
It is necessary and not sufficient, and we have the data to say so. We enriched 72 IT firms our research had already measured as missing from AI answers, and 28 of them ran their websites on HubSpot CMS, meaning clean markup and total control of what a crawler sees. They were still absent. Your own pages tell an engine what you claim about yourself. The other five citation types tell it whether anyone else agrees, and that second question is the one that decides a recommendation.
Sources and further reading
- Profound: brands across four or more platforms are about 2.8 times more likely to be cited; Reddit is roughly 12 percent of ChatGPT citations and Clutch roughly 2.4 percent.
- Kevin Indig with G2: G2 holds about 22.4 percent share of voice for software, and roughly 10 percent more reviews is associated with about 2 percent more AI citations.
- SE Ranking 2026: structured data appears on 71 percent of ChatGPT-cited pages and 65 percent for Google AI Mode.
- AI keeps recommending the big names: our fifteen-region study, where 150 of 228 Clutch-listed firms were named by no engine.
- They run HubSpot and buy Google Ads: the 72 enriched firms, 28 of them on HubSpot CMS and still absent.