Hi Reader,
For three years, our industry argued about whether AI licensing deals actually change what the models cite.
This week somebody finally counted.
129 million citations. 91 confirmed deals. Publishers with an OpenAI agreement earn 48% more citations in ChatGPT than those without.
And that's just the headline.
Across 16 US industries, the outlets carrying the majority of AI citations in 15 of them were niche and trade publications. Unlicensed. Unglamorous. Pulling 213% more AI citations than mainstream media.
So the outlet you've been chasing for six years may be worth less to a model than the trade pub your client keeps calling "small." You don't know which is which until you look, and most teams have never looked.
ππ½ Here's the prompt:
π― Expert Mode: The "Citation Map"
Scenario: You're about to build next quarter's media list off habit, legacy relationships, and a spreadsheet somebody made in 2023. Meanwhile, your buyers are asking ChatGPT about your category and getting answers assembled from a completely different set of sources. You need to know which outlets the models actually reach for before you spend another quarter pitching the ones they ignore.
Instructions:
1οΈβ£ Write down 8 to 12 real questions someone would ask an AI about your category. Not brand questions. Buyer questions. "Best project management software for nonprofits." "Is this treatment covered by insurance?" "Which firms handle this kind of case?" The things your prospects actually type.
2οΈβ£ Run every question through Perplexity, then repeat a handful in ChatGPT with search on. Copy the cited sources for each answer into a single document, and note which domains keep appearing. This takes about 20 minutes, and it's the whole ballgame.
3οΈβ£ Paste that source list into Claude alongside your current target media list, then run this:
Act as a media strategist auditing whether our earned media targets match what AI engines actually cite in our category. Below are two lists. First, the sources cited by AI answer engines across a set of real buyer questions, with the question each source appeared under. Second, our current target media list.
β
Build a Citation Map with five columns: (1) outlet, (2) how many of our buyer questions it was cited under, (3) whether it's currently on our target list, (4) what content format earned the citation, such as a buying guide, a ranked list, original research, or a service explainer, and (5) the single most realistic angle we could pitch that outlet given what it already gets cited for.
β
Then give me three lists. Outlets that earn citations and are missing from our target list, ranked by frequency. Outlets on our target list that never appeared in the citations at all. And content formats we don't currently produce that are earning citations for everyone else.
β
Do not pad the map with outlets that weren't in the source data. If our target list is mostly absent from the citations, say so plainly.
[PASTE CITATION SOURCES AND TARGET MEDIA LIST]
4οΈβ£ Read the output looking for one specific thing: the gap between prestige and citation. The outlets that make your CEO happy are often not the outlets that make the model confident. Both can be true, and now you can price the difference.
5οΈβ£ Take the top five cited outlets you aren't currently working, give each one an owner and one pitch angle, and get the first outreach out this week. Rerun the map in 90 days and see what moved.
Suggested Tool: Claude (holds two conflicting lists side by side without quietly inventing a third)
Why You'll Love It
- It replaces "who do we want coverage in" with "who does the machine trust," which is a much harder question to argue with in a planning meeting.
- It gives your trade pitches a real business case, so the client who dismisses the vertical outlet can see exactly what it's worth in AI answers.
- It exposes the formats you aren't producing, because the citations are going to buying guides, ranked lists, and original research, not to your quarterly announcement.
The brands that win the answer layer won't be the ones with the best clip book. They'll be the ones a model reaches for by default. Map that first, then go earn it.
Talk soon,
Chris
β‘ The Cowork Version
Same map, except Claude runs the citation capture itself, reads your actual media list, and leaves a finished audit in your folder.
Setup (one time):
- Make a folder for this and drop in your current target media list however it exists. A spreadsheet, a CSV export, or a plain text list all work.
- Add a text file with your 8 to 12 real buyer questions, one per line.
- Optional: export your last two quarters of earned coverage so Claude can compare where you actually landed against where the citations go.
- Optional: connect Google Drive or Notion if the media list already lives there.
Run this Cowork command:
Build a Citation Map from the files in this folder and save the outputs here.
β
1οΈβ£ Read my buyer questions file. For each question, search the web and capture which sources are cited or surfaced as authoritative. Record the outlet, the URL, and the page format, such as a buying guide, ranked list, original research, or service explainer.
β
2οΈβ£ Read my target media list and normalize it to domains so it can be compared against the citation data.
β
3οΈβ£ Build the Citation Map: outlet, citation frequency across my questions, whether it's on my target list, the format that earned the citation, and the single most realistic pitch angle given what that outlet already gets cited for.
β
4οΈβ£ Produce three gap lists. Cited outlets missing from my target list, ranked by frequency. Target list outlets that never appeared in any citation. And content formats that earn citations I don't currently produce. Do not add outlets that weren't in the captured data.
β
5οΈβ£ If a coverage archive is present, compare where we actually earned placements over the last two quarters against where the citations go, and tell me what percentage of our earned coverage sits in outlets the models never cite.
β
6οΈβ£ Save the map as Citation Map.xlsx and save a one-page brief as Citation Map - Pitch Brief.md covering the top five uncovered outlets with owner, angle, and format.
What you'll get:
- A
Citation Map.xlsx ranking every cited outlet in your category by how often the models reach for it
- A gap list of high-citation outlets you aren't pitching, which is your new target list
- A dead-weight list of legacy targets earning you nothing in the answer layer
- A format gap showing which content types are winning citations while you keep shipping announcements
- A
Citation Map - Pitch Brief.md you can hand to the team Monday without reformatting
Want it on autopilot? Ask Claude to save this as a skill and rerun it quarterly against the same question set. The map becomes a trend line, and the trend line is the slide that gets your GEO budget approved.
Here are a few articles from the past week to help keep you up to speed:
1. Publishers With OpenAI Deals Earn 48% More AI Citations on ChatGPTβ
GlobeNewswire (Press Ranger and OtterlyAI)
August 20, 2026
The first study to pair large-scale citation monitoring with a verified record of AI licensing deals: 129.3 million citations across seven platforms, matched against 91 confirmed agreements. Publishers with an OpenAI deal earn 48% more citations on ChatGPT. But the finding that matters for your media list sits lower in the release, where niche and trade outlets, most of them unlicensed, carried the majority of news citations in 15 of 16 US industries and pulled 213% more citations than mainstream media. This is the story behind this week's prompt.
2. Creators are moving into brands' AI search playbooksβ
Digiday
August 20, 2026
AI visibility is now a line item on the creator brief. Agencies are auditing which creators already earn citations in LLM answers and buying against that data, while asking creators to write captions machine-readable enough to get scraped. One caution worth carrying into your own planning: volume is not citation. As one exec put it, 300,000 TikToks in the wrong place get ingested by nothing.
3. Perspective: In the AI Era, Relevance is a Communications Outcomeβ
PRovoke Media
August 19, 2026
Margot Edelman argues that as AI raises new questions about what companies are actually for, relevance stops being a marketing feeling and becomes a communications deliverable with a number attached. Useful framing for the budget conversation, because relevance can be measured in a way awareness never could.
4. OpenAI's ads business hits Europe at the six month markβ
Digiday
August 19, 2026
ChatGPT ads go live across 31 European markets on August 24, bought through Publicis, Omnicom, WPP, Havas, Dentsu and MediaPlus. Six months from a US pilot to near-global rollout, with 900 million weekly users and roughly 20% of queries showing commercial intent. The answer layer is now an ad-supported channel, which means the citations you earn there are sharing space with the ones somebody paid for.
5. PR AI Update: What Happens to the Story When AI Sits in the Middle?β
PRovoke Media
August 18, 2026
Evan Zimmer digs into the mechanics of mediation: how models select, organize, and synthesize information, and how that process quietly changes the story. Worth reading right before you build your citation map, because it explains why the source a model picks matters more than the sentence you wrote.
β