Track PR Wins in AI Citation Data Using Source Intelligence

A media list ranks outlets by reach and domain authority, not by which of them an AI engine pulls from in your category. This workflow starts from the sources already cited in your answers, separates those carrying a listed placement offer from those you must earn, and baselines both before the first pitch.

The GEO problem it solves

Chasing the biggest names first misreads how AI answers get built. They draw on third-party editorial sources unevenly, and the mix shifts by provider, prompt and category. A trade publication recurring across your monitored answers can matter more than a famous outlet that never appears.

Citation evidence also changes what you do about the control gap. You cannot edit a competitor roundup, but you can see which URLs carry the citations, which competitors co-occur there while you are absent, and pitch the format that outlet gets cited for.

How the workflow breaks down

1. Map the cited landscape. get_source_domains returns the domains cited inside the requested evidence scope with type, mention count and average position. Filter to MEDIA, then BLOG and INSTITUTIONAL, using the same window and optional provider, topic, tag or funnel filters throughout. It defaults to 30 days and returns 50 rows by default, 100 at most.

2. Profile the gatekeepers. get_source_profile returns a domain's mentions, providers, top cited URLs and the SELF and DIRECT competitors co-cited in the same answers, over the last 30 days unless you widen it. It does not return the prompts behind the citation, and its top URLs tell you which format earns citations there.

3. Split paid from earned. get_backlink_opportunities returns, already ranked, the cited domains carrying an available listed placement offer, with the mention count, answers citing it that name a direct competitor, answers naming you, a priority label, a domain rating, the platforms and the cheapest published price. very-high means cited at least twice, a direct competitor named in answers citing it, you never named; medium means the opposite. Social domains, your own and your competitors' are excluded; everything absent is earned-only.

4. Freeze the baseline. get_brand_performance gives mention rate, source rate, share of voice and sentiment. Source rate is the share of answers citing any source that cite your own domain, not the share of your pages cited. get_perception_history adds monthly sentiment and alignment scores.

What a run actually returns

A valid run separates pitchable editorial domains from social platforms, marketplaces and your own properties inside one dated evidence scope. Each row carries citation volume and average position, while the profile adds URLs and competitor overlap.

The gap between the highest-volume source and the first pitchable source is the finding. A media list built on reach can invert that order. Start outreach at the highest-ranked pitchable domain where the scoped evidence shows competitor overlap and your brand is absent, not at an undated global total.

What to do with the output

Work the very-high rows top to bottom in the order returned, then the high rows. Stop at medium: there you are already named and no competitor is, so a placement buys you nothing. A missing price means the offer exists but is unpublished: a call to make, not a row to skip. A listed placement is a purchase, not a citation: you buy presence on a page an engine already reads.

If that call returns no rows, no cited domain here carries a listed offer. Nothing is broken, the whole list is earned: rank the profiled domains by direct-competitor co-citations, highest first, and pitch each the format its top cited URLs already use.

Movement after a placement is not lift. Publication, crawl access, indexing and retrieval sit between the two, and an unchanged profile a month later is not proof of failure. Re-run identical snapshots monthly so the windows stay comparable.

The Complete Prompt

I want a PR target list built from AI citation data. List the brands I monitor and, if there is more than one, ask which to use before pulling data.

Walk me through this step by step, and explain what each step reveals before moving on:

1. Pull source domains for the last 30 days, filtered on MEDIA, then BLOG, then INSTITUTIONAL, and order each by mention count. Preserve any provider, topic, tag or funnel scope for every evidence call below.
2. Run source profiles on the top five editorial domains over the same scope. Show mentions, top cited URLs, providers and the SELF and DIRECT co-citations. Do not claim it maps sources back to prompts; it does not.
3. Pull backlink opportunities over the same scope. Per row show the domain, mention count, answers citing it that name a direct competitor, answers naming me, the priority label, the domain rating, the platforms and the cheapest published price.
4. Build the target list in two blocks. Paid block: the backlink rows in the order returned, very-high first then high, dropping medium and below. Earned block: every profiled domain with no offer, ordered by direct-competitor co-citations, each with a pitch type, inclusion in an existing article or a new angle, tied to its top cited URLs.
5. Record the baseline before any pitch: brand performance over the same 30-day scope, perception history and a source-profile snapshot per target. Re-run monthly on identical filters and report movement as correlation, not lift.

If backlink opportunities returns no rows, say that no cited domain here has a listed offer, skip the paid block and build the earned block alone. If a source profile returns no co-cited competitors, keep the domain and mark its gap unproven rather than dropping it.

Ask me one thing before the final list: which of these publications I already have a relationship with, since that changes the pitch type and nothing in the data knows it.