How to Get Cited by AI When 85% of Citations Aren't Yours
TL;DR
- Even top-quartile brands own only about 15% of the citations behind AI answers about them, so third-party sources need management and a plan.
- Third-party pages are also where AI picks up outdated descriptions of your brand, and correcting existing coverage is a fast, low-effort way to improve accuracy.
- Your website plays a part too, so don’t sleep on its importance. Third parties paraphrase it, and buyers visit it to confirm what an AI model told them.
Part 2: How to Get Cited by AI When 85% of Citations Aren't Yours
Read Part 1: AEO Strategy: The Organizing Principle, Not a Bolt-On
One of Edgar Allan's clients, Satellite Industries, is the most-cited brand in the portable restroom category across ChatGPT and Copilot, yet its own website holds just 11 to 15% of the AI citations in that category (14.1% in September and 10.7% in October 2026). It’s a pretty good example of the divide between first and third-party citation source weight: if your plan to get cited by AI only optimizes your own website, it's working on the smallest slice of the signal, and your best bet is to fund a managed third-party program alongside optimizing your site.
Zane Coffin, our Senior Manager of Brand Strategy and Performance, brought back a similar number from Profound’s Partner Kickoff New York: even top-quartile brands own only about 15% of the citations behind AI answers about them. The two figures measure slightly different things. Satellite’s is its share of every citation in its category, and Profound’s is a brand’s share of the citations in answers about that brand.
Both say the same thing, though: most of your signal lives off of your site.
Most of what AI says about you lives on someone else’s site
In Profound's tracking, satelliteindustries.com has about twice the citation share of the next competitor. Here's where the category's citations came from between October 1 and 29, 2026:
Source: Profound, Portable Restrooms category, ChatGPT and Copilot, October 1 to 29, 2026 (snapshot). Percentages are shares of all citations in the category.
Portable restrooms may be one of the less glamorous (if not most vital) categories we track, which makes it a useful one: nobody writes a buyer's guide about them for prestige, so the citation map is about as unvarnished as they come. A Walmart product page, a competitor's comparison, and a trade magazine's buyer guide each say something about Satellite, and a model treats all of them as evidence. At Edgar Allan, our own category looks much the same: about 95% of web agency citations across ChatGPT and Copilot in October 2026 went to sites other than edgarallan.com. That's our share of the whole category rather than of citations about us, so it's a different measure from the 15%, with the same lesson.
Every trade article, retail listing, and association profile paraphrases your website, so none of them can describe you more clearly than your site does. Call it the paraphrase ceiling: your site language sets the upper limit on how well anyone else can describe you, so it should be as accurate, differentiated and high fidelity as you can make it. Getting cited well means correcting what third parties already say, giving them something new and specific to repeat, and keeping the source they all copy from clear enough to copy well.
Unmanaged sources describe an old version of you
Models don't date-stamp what they read. So, a 2022 trade article and last week's homepage edit can sit side by side in one answer, with nothing telling the buyer which one is four years old. AI sentiment about a brand is typically built from 50 to 200 citation sources, and the ones you’re not managing and just letting age out there on the internet are the likeliest to predate your last rebrand, pricing change, or product launch.
We covered a clean example after Zero Click NY. Southwest Airlines moved to assigned seating, yet answers kept describing its famous open boarding, because most of what the web had written about the airline predated the change. Southwest's own site was right the whole time. The model was summarizing everyone else.
Zane calls the fix the short game: keep working relationships with the publications that already cover your category, so you can get a post corrected, updated, or filled in with something new you've published. For Satellite, that means the trade publications promonthly.com and pumper.com and the Portable Sanitation Association International at psai.org, which together carry more than 11% of the category's citations, roughly as much as Satellite's own site.
Favor the editors who cover your category every month. G2 shared at Zero Click NY that fewer than 3% of the data influencing AI answers comes from paywalled tier-one publications, so a corrected buyer's guide in your trade press usually moves more than a splashy feature behind a paywall. Paid "best of" placements and high-volume Reddit posting fall outside this work, since those mentions say little and inherit the volatility of whatever platform hosts them.
Off-site share can vanish in a week
Third-party citation share moves in ways no brand controls. Zane's example is Reddit, which "was a really big source of citations" for reviews and sentiment, "and then that just fell off a cliff" (we covered the numbers in Part 1). Google's AI Mode shows a second kind of shift: Zane has watched it start listing far fewer brands per answer, with no matching rise in third-party citations, so every brand in a category is competing for fewer slots.
The answer is to spread the work across source types (trade press, associations, retail listings, customer content) so a shift on any one platform costs you a just slice of visibility rather than all of it, and to anchor the whole thing in your own site. A program that depends on a single off-site platform is as fragile as one that ignores off-site sources entirely.
Every third-party fix connects back to your website
Zane's point is that most citations live off-site. I'd take it a step further: that's exactly why your focus on and budget for your website should stay put. As citations spread across the web, the site actually matters more, because it's the hub that holds what Zane calls the deepest version of the brand's truth.
Every trade article, retail listing, and association profile paraphrases something, and the best version of that something should be on your site. When Satellite asks pumper.com to update a buyer's guide, the first place that editor will check the claim is satelliteindustries.com. That's the paraphrase ceiling at work: if the brand differentiators and/or story can't be summarized accurately because they’re muddy or inconsistent, every corrected post is paraphrasing the same mud.
So my advice is to split the budget rather than move it. Maintain the website for messaging, structure, and conversion, and add a line item for the third-party program alongside it, sized by the diagnostic below.
The long game is numbers nobody else has
Once corrections clean up what's already written, the long game is giving the web something new to say. Zane describes models as "very thirsty for information," and the generic version of any topic pretty much already exists somewhere else, so a model has no reason to cite yours without your detail and narrative. For example, concrete numbers only you can publish give trade editors something to quote and give models a source they can't get anywhere else. The next post in this series makes the full case for how to build that: the Content Lab model.
Start Q1 with a diagnostic you can run in an afternoon
This diagnostic shows how much of your category's citation share sits in sources you can influence, and that sets the split between owned and earned work. Here's what a first quarter looks like.
- A diagnostic. Run buyer-style prompts across ChatGPT, Gemini, Claude, and Perplexity, and record whether you're named, how you're described, and which sources get cited. It takes hours, and the cited sources become your third-party target list. Our AEO playbook covers how to pick the prompts.
- A correction list. Rank the off-site pages models cite most, check each against your current positioning, and request updates, starting with trade media and industry associations.
- One original-data piece. Publish one set of numbers nobody else has and pitch it to the editors on your correction list.
- Monthly measurement. At minimum, run manual checks. A tracking platform is even better, with continuous monitoring, and using something like Profound's Fact Check can compare what AI systems say about you against what you've said about yourself. Our free AI Discovery Health Check is a quick way to set a baseline.
- A named owner. Put someone above the SEO function in charge, with reach across brand, PR, content, and web.
If the diagnostic shows models describing you vaguely, tighten your site's brand positioning before sending any correction requests, since every request points an editor back to it.
This is how we approach AEO at Edgar Allan: as part of brand-to-build engagements built on Visibility & Optimization, where brand, SEO, AEO, and CRO run as one loop, and the story, the site, and the signals models pick up are planned and executed on together. If you want help mapping where your citations come from, let's start a project.
This is your signal to fix what the web already says about you, give it something new worth repeating, and make sure the page it sends buyers to backs it all up.
This article is the second in a series that grew out of a conversation between Edgar Allan founder Mason Poe and Zane Coffin, Senior Manager of Brand Strategy and Performance at Edgar Allan, for Building the Next Web. It's the same series that's brought us conversations with Finsweet’s Alex Iglesias, ourCMO's Josh Webb, Code & Wander's Alessia Sannazzaro, Edgar Allan's own Witt Langstaff, Webflow's Nathan Huening, and more.
FAQs
How do I get cited by AI if most citations come from other websites?
Start by finding out which sites models already cite when buyers ask about your category, then work on those pages directly. Correct outdated or inaccurate coverage, keep retail and directory listings current, and publish original data that trade editors and associations want to reference. Keep your own site clear and specific, because every third party that writes about you will paraphrase it. Brands that get cited consistently manage all three at once.
How can I increase AI citations without paying for list placements?
The most reliable route is earning mentions that carry real information. Build relationships with the trade publications and associations that already cover your category, and give them something worth quoting, like proprietary numbers from your own work. Paid "best of" placements tend to say little about you and can lose value quickly when platforms change how they weight sources.
Should we cut our website budget to fund PR and earned media for AEO?
No. Third-party sites paraphrase your website, so cutting it weakens the source every correction and new mention depends on. Buyers also land there to check what a model told them. Add a third-party line next to the website budget, sized by how much of your category's citation share sits in sources you can influence.
Which third-party sites should we work on first?
Start with the domains models cite most often in your category, then sort them by how much influence you have. Trade publications and industry associations come first, because they cover the category regularly and their editors are used to hearing from the brands they write about. Retail and directory listings come next, since you can often edit them yourself. Competitor pages and general review sites are the hardest to change, so the best response there is a clearer site and stronger original data.
Do AI brand mentions on Reddit and social media still matter?
They can, but they're the least stable part of the mix. Reddit's share of AI citations dropped sharply this year, and in Satellite Industries' category, social platforms account for just 0.3% of citations. Treat social and community mentions as one input among many, and put steadier effort into trade media, associations, and your own site.
Can we ask a publication to change what it wrote about us?
Yes, and it works best with outlets you already have a relationship with. Ask for one of three things: a correction when a detail is wrong, an update when the post describes an old version of your product or positioning, or an addition when you've published something new the post is missing. Bring a specific source the editor can check, usually a page on your site. Keep a running list of requests, and track which cited pages change and whether AI descriptions of your brand follow.