How to earn citations on the sources AI actually reads

The uncomfortable finding in AI visibility work is that your website is not where the decision gets made. When an engine answers "what's the best tool for X", it mostly reads pages published by other people. If those pages don't mention you, you aren't in the answer, no matter how good your own site is.

So the work becomes: find those pages, and get into them honestly. This is a method for doing that.

Step one: find out what's actually being cited

Don't guess. Run the buying questions your customers ask through the engines and record which URLs come back as citations. Perplexity shows them by default. ChatGPT shows them when it retrieves. Do this across several phrasings ("best X", "X alternatives", "X for small teams") because different phrasings pull different sources.

What you'll get is a list, usually shorter than expected. Most categories have somewhere between ten and thirty sources that engines return over and over. That list is the entire playing field.

Sort it into three buckets:

• Sources that already mention you. Verify the description is accurate and current. A stale entry is worse than none; it teaches the model something wrong about you.
• Sources that mention competitors but not you. This is the target list. Everything below is about this bucket.
• Sources you can't realistically enter. Some publications have editorial processes you don't fit, or cover a segment you're not in. Being honest about this saves months.

Step two: understand what each source wants

Sources fall into rough types, and each has a different, legitimate entry path.

Review platforms (G2, Capterra, TrustRadius and their category-specific equivalents) want a claimed, complete, accurate listing and real reviews from real customers. The listing you control entirely: claim it, fill every field, keep it current. The reviews you influence only by asking customers who've had a good experience, at a moment when that's true, without incentivising the content of what they say. Incentivised reviews are both against most platform terms and detectable.

Category roundups and listicles are written by someone. That someone usually has a public byline and a stated methodology. The legitimate path is the boring one: find the author, read what they cover, and send a short note explaining what you do, who you're for, and why you're a fit for that specific piece. Include what makes you different in one sentence. Most get ignored; some don't. This is digital PR, and it works at the rate PR works.

Community threads (Reddit, Hacker News, industry Slacks and forums) are where a lot of honest comparison lives, and where the most damage gets done by people who try to game it. The only approach that survives is participating as yourself, disclosing your affiliation, and being useful in threads whether or not they're about you. Communities are extremely good at spotting the alternative, and the penalty is durable.

Independent analyses and comparison sites vary. Some accept vendor submissions. Some are one person with strong opinions. Read a few pieces before reaching out and you'll know which.

Step three: make yourself easy to include

Before the outreach, remove the friction. When someone decides to write about your category, the vendors that make it easy to be described accurately get described accurately.

That means a page on your own site that states plainly: what the product does, who it's for, what it costs, and how it differs from the obvious alternatives. Not marketing language: the plain version a journalist could quote. Add structured data so the facts are machine-readable. Keep pricing public if you possibly can; "contact us for pricing" gets you excluded from comparison pieces more often than people realise, because the writer can't fill the column.

This is the part you fully control, it takes an afternoon, and it compounds against every other step.

Step four: expect it to take time

Here's the sequencing that trips people up. Getting into a source is step one. That source then needs to be re-crawled. The engine's retrieval needs to start surfacing it for the queries you care about. Only then does the answer change.

That chain takes weeks at minimum, and often longer. If you fix something and re-check the next day, you will see nothing, conclude it didn't work, and stop right before it would have started working. Track on a schedule and judge on the trend.

What to avoid, and why it fails on its own terms

The shortcuts are well known: paid placements dressed as editorial, review campaigns with incentives attached, spun articles across low-quality domains, sockpuppet accounts in communities.

Set the ethics aside for a moment and look at the mechanics. The sources that engines retrieve most heavily are the ones with the strongest filtering: editorial review, community moderation, platform fraud detection. Those filters are the reason those sources get trusted and retrieved in the first place. A tactic that only works on unfiltered sources places you on exactly the pages engines weight least. You end up paying for mentions that don't move the number.

And when the filter does catch it, the cost isn't a lost placement. It's a source that now describes you as the vendor that tried it.

A reasonable first month

If you want something concrete to run:

1. Week one. Record citations across ten buying questions and five AI surfaces. Build the source list. Sort into the three buckets.
2. Week two. Fix everything on your own property: the plain-language page, structured data, public pricing, accurate existing listings.
3. Week three. Claim and complete every review-platform listing. Start asking recent happy customers for reviews, with no incentive attached.
4. Week four. Pick the five highest-value roundups that name competitors but not you. Find the authors. Send five specific, short, honest notes.

Then wait, and re-measure monthly. The first month is mostly setup; the movement shows up later. That's not a sales caveat, it's just how the mechanism works.