Find the sources AI engines cite, then rank them for outreach

DeepSmith frames the shift this way: getting named in a ChatGPT or Perplexity answer is usually not about your own site alone, but about the third-party domains the engines trust and pull from. Once you can see that list, outreach becomes a targeted plan aimed at review sites, listicles and roundups the engines already lean on.

Semrush gives a reason to care beyond links. Its study with Kevin Indig found that 62% of AI citations do not result in a brand mention. A page can be cited without naming the product you want named, so getting your brand onto a page that engines already cite is worth more than another page of your own.

Semrush also points out that cited pages can be wrong. In its example, Claude repeated an outdated claim about a product from a review, even though the product documentation said otherwise. Part of outreach is correcting what a cited page says about you.

It is long and flat. On Linkeddit's test project, the five most-cited domains in ChatGPT's answers hold about 21% of citations: matomo.org 10 of 183, itechguides.com 9, the project's own plausible.io 8 and analytics-alternatives.com 5. The remaining 79% sits in a long tail of other domains, so a list of just the top few sites misses most of the opportunity.

The engines differ. In Google AI Overviews the leading cited domains were analytics-alternatives.com and youtube.com at 6 of 107 each, with reddit.com at 4. A page that one engine cites often may not appear for another, so keep the engines separate when you rank pages.

Two of the top ChatGPT domains, itechguides.com and analytics-alternatives.com, are neither the customer nor a tracked competitor. Those are the outreach candidates. The gap record gives the question to attach them to: an open gap where the brand was neither mentioned nor cited in 3 successful samples.

Semrush lists six factors: citation impact, prompt relevance, competitor presence, the size of the brand gap on the page, publisher authority and outreach feasibility. It adds that competitor pages and some government sites are less receptive. Do not exclude competitor domains blindly, though, since a competitor's comparison page can still be worth pitching.

Split the factors across the connectors. Linkeddit gives you citation impact per engine and the question each page answers. Ahrefs gives you publisher authority and backlink context through tools such as site-explorer-domain-rating and site-explorer-referring-domains, plus brand-radar-cited-pages for how often a page is cited.

Semrush recommends a human reviews each opportunity, because a publisher who finds an error in a pitch may ignore it, and suggests asking for a low-effort change instead of a long list of additions. Claude drafts, you decide, and nothing is sent automatically.

The ranking pages for this query are written around one vendor's product. Semrush's workflow starts from a CSV export of its own prompt tracking, and DeepSmith's page compares five tools, including Ahrefs Brand Radar, whose AI Citations report it describes as the core screen for this job. None shows how to combine the measured citation data you already have with a separate authority source in one conversation.

This workflow also keeps the question attached to the page. Because Linkeddit's gap record names the buying question where the engine recommended someone else, Claude can write a pitch that says why the page matters for that question, not only that it is popular. The limit is that citation data is a controlled measurement proxy, not a copy of what every user sees, so check a page before you pitch it.