Citation Gap Analysis: Find and Pitch the Pages AI Cites for Rivals

Step 1: List the Answers That Left You Out

Start with about 20 questions your buyers ask. Our guide to prompt research covers where to find them. Ask each one on each engine you care about, at least twice, because answers change between asks. A rival is any company in the answer that sells what you sell. Keep one row per link, in a sheet with these columns:

Date, Question, Engine, Named you?, Rivals named, Link, Where it goes, What the page says.

Now filter to the answers that named a rival and not you. In our sample, that was about half of answers. The links in those rows are your raw list.

Step 2: Strike What You Can't Pitch

You can do this from the address and the name the site gives itself, before reading the page. With the few links to your own site, it removes 47.5 of every 100 links. Each kind you strike still has its own play.

The rival's own site

This is the largest single strike, at 33.4 in 100. A rival's "best tools" article can look like any other list, but it won't add you. Answer it with a page of your own on the same question, and run that page through our 12 checks for a list on your own site.

Forums and communities

At 12 in 100, there's no editor to write to. Join the conversation under the community's rules and say who you are. For Reddit, work the subreddit rather than the thread.

Social and video

These are 1.8 in 100. Post your own, or work with a creator who already covers your category, and follow the disclosure rules in Step 5.

Step 3: Open Every Page That's Left

What remains are other people's pages. Open each one and search it for your name and each rival's name. It's quick, and it sorts them three ways:

Why a link isn't enough

A link beside a claim doesn't prove the page says it. A 2023 study of four engines of that time found that only 74.5% of citations supported their sentence. The Tow Center found "a common tendency to cite the wrong article" across eight tools asked to identify news articles from excerpts. A study from Washington University in St. Louis of Google AI Overviews, where a model judged the claims, found 11.0% unsupported by the cited pages, or roughly 5.3% on its most generous assumption. None of them tests brand names, which is what we added. When we looked through a sample of the pages naming neither by hand, many were how-to guides, other vendors' product pages and public-sector pages.

Step 4: Rank What Survives

You now have pages that name a rival and not you. Put these three questions to each one, in this order, and pitch your top ten first.

Does the rival have an entry?

A name in a passing sentence isn't a place on a list. Of the pages that named a rival, 68.8% named it in a heading. Counted by links, 21.7 of every 100 land on a page that names a rival in a heading and doesn't name you. Start there.

What kind of page is it?

Lists and reviews are far more likely to be real gaps than guides are. A rival was in a heading on 85.9% of the ranked lists naming one, against 41.4% of the how-to guides.

Does it keep coming back?

Most third-party pages cited in these answers are one-offs: 81.5% were cited in just one answer that left the company out, and 5.2% appeared on more than one engine. A page that shows up again sits in more of the answers missing you, so try it first. Don't expect a short list to close the whole gap, though: a company's ten most-cited pages carried 19.5% of its links to third-party pages on average.

Is the Pitch Worth Sending?

We haven't tested what happens to AI answers after a page adds a brand, and we found no neutral study that has. The closest we can offer is a comparison inside single companies. When a cited page named the company, the answer did too 89.4% of the time. When it didn't, 70.5%.

Comparing inside a company takes brand size out of it, but not the differences between questions. The same reasons that put a company on a page may also get it named in the answer. The comparison also rests on a handful of companies, mostly ones AI already names often, and not all of them point the same way. For more than half of the companies, no page in our fixed sample named them at all, so there was nothing to compare.

So read it as a reason to try, not a promise. OpenAI makes none: it says of ChatGPT search that "Placement is not guaranteed". Google also warns that "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem." A place you earned on a page people read still reaches those people.

Step 5: Send the Pitch

Who to write to

On pages that named a rival and not the company, 75.8% declared an author in their HTML on average, and 76.3% linked to a contact page or an email address. Write to the named person first, and use the site's contact route if there isn't one. Then check the date: 77.9% declared one, and 64.9% one from the last 12 months. A page nobody has touched in years is less likely to change for you.

What to send

Keep it short, name the page, and hand over an entry they can use as written. Our guide to earning a spot on lists covers how to write that entry. Here's a template to fill in:

Paid places and affiliate lists

On the same pages, 15.5% on average used words like "affiliate" or "sponsored". That's a rough text match, but it suggests some may answer with a price or a commission program. Before you pay, know the rules. This is FTC guidance and Google policy, not legal advice:

• The FTC's Endorsement Guides say a connection that could affect how an endorsement is weighed, and that the audience wouldn't expect, "must be disclosed clearly and conspicuously". That covers anyone paid or given something of value to mention your product, and free products count.
• Google's spam policies list "buying or selling links for ranking purposes" as link spam, including sending someone a product in exchange for a post with a link. Paid links are allowed when they're marked as sponsored with rel="sponsored" or nofollow.
• None of it buys a place in an answer. A paid entry is one more page a model may or may not read.

Step 6: Check It Worked

First open the page and confirm your entry is live and accurate. Then ask the same question on the same engine each week and watch two things: whether the page is still cited, and whether the answer names you.

Don't count one appearance as a win. In our tracking before the October 2026 upgrade, where a run is one pass through the questions, a new mention was still there on the next run 54.3% of the time, and one seen in two runs in a row 96.1%. So wait for two in a row. After four checks with no mention, move on to the next page. We can't tell you how long a change takes to show, because we have no data on how quickly engines pick up an edited page.

What This Can't Tell You

• Pages change. We read each page once, after the answers that cited it. A page may have said something else at the time.
• We can't see why a name was in an answer. A page naming the rival isn't proof the rival's name came from it. It could have come from another link, or from what the model already knew.
• Names can be missed. We matched whole names in the text. A different spelling or a parent company's name reads as no match, so "names neither" may run a little high.
• Our setup shapes the mix. These answers came through the engines' APIs with instructions of our own, so forums and review pages may take a larger share than in a plain chat.
• Most answers predate the October 2026 model upgrade. About two-thirds do. After it, rivals' own sites read about the same (32.3% before, 33.3% after), forums took more (11.8%, 15.4%) and third-party pages less (53.8%, 48.5%). Few companies sit on both sides, so that's a sign rather than a test.

How We Measured

We took the answers from Ranqo's tracking, then fetched the pages ourselves and read plain facts from their HTML, with no AI model involved:

How Ranqo Fits

Ranqo does most of the first two steps for you. It asks your tracked questions on your engines every week, and the Citation Pool tab on the Sources page shows, site by site, which brands AI names in the answers citing it and how many of those answers left you out.

The Publications lane in Outreach turns that into a ranked queue of lists, comparisons, reviews and alternatives pages, leaving out your own site and your competitors'. Each card opens the page for the check in Step 3 and drafts a pitch for you to edit and send yourself, because Ranqo never sends anything. On Head-to-head, you can see the pages and threads from Outreach for one rival.

If a target serves a recommendation still open in the Action Center, marking it done moves that recommendation to Done and starts counting runs. When you take a reading, an answer counts as won only if the last two runs named you, and a flat reading shows Too early to call until enough runs have passed: 4 for outreach work.