Hi Reader,
Two studies landed within a day of each other this month.
The first says earned media is the biggest player in PESO for AI discoverability. In a survey of 168 PR executives, 44% ranked earned first, ahead of paid, owned and shared. More than half now put most of their AI discoverability budget behind it.
The second analyzed 7,387 AI citations across ChatGPT, Perplexity, Claude and Gemini. Brand-owned content took 88.3% of everything cited. Product pages alone got 24.1%. Reddit and YouTube combined got 4.2%.
Both are real. Both have numbers. They point in opposite directions.
Last week we stress-tested other people's stats. This week, the problem is simpler and worse: neither study is about your company. One surveyed what PR people believe. The other measured one agency's B2B SaaS client base at a single point in time.
Your CFO won't accept either one. Nor should they.
ππ½ Here's the prompt:
π― Expert Mode: The "Earned Media Budget Defense"
Scenario: Budget season. Someone above you is going to ask why earned media deserves a line in the AI visibility budget when the owned-content vendor has a deck full of charts. You have industry studies that contradict each other, and none of them looked at your category, your buyers, or your brand. You need a number you generated yourself.
Instructions:
1οΈβ£ Write down 15 to 20 questions a real buyer would ask an AI assistant when they're close to choosing. Not "what is [category]." More like "[your company] vs [competitor] for [use case]" and "how does [your company] handle [the thing procurement asks about]." These are the moments where money changes hands.
2οΈβ£ Run all of them across at least three models. Record every source cited in the answer: the URL, whether it's yours or someone else's, and what type of page it is. Owned site, product page, earned coverage, directory listing, forum, analyst note. This takes about 40 minutes and it is the only part nobody can do for you.
3οΈβ£ Pull your last 12 months of coverage into a simple list, and your current AI visibility spend broken out by line. Then paste all three inputs into Claude and run this:
Act as a finance partner to a communications team preparing a budget defense. I'm going to give you three things: a log of what AI models actually cited when asked buyer-stage questions about us, a list of our earned coverage from the last 12 months, and our current AI visibility spend by line item.
β
First, tell me what the citation log actually shows. What percentage of citations came from sources we own, sources we earned, and sources we neither own nor earned? Show your counts, not just percentages.
β
Second, cross-reference the earned coverage list against the citation log. Which specific placements are showing up in AI answers, which are invisible, and is there any pattern in outlet type, recency, or format that separates the two?
β
Third, tell me honestly where my sample is too small to support a claim. If I ran 18 prompts, say plainly what 18 prompts can and cannot prove, and what I would need to run to make it defensible.
β
Then draft a one-page budget recommendation with a specific allocation, the reasoning behind each line, and the single strongest counterargument someone in that meeting will raise. Write the counterargument as fairly as you can, then answer it.
β
Do not use any industry benchmark or third-party statistic unless I provide it. This case has to stand on my own data.
[PASTE CITATION LOG, COVERAGE LIST, AND CURRENT SPEND]
4οΈβ£ Read the cross-reference section first. The interesting finding is almost never the headline percentage. It's discovering that the trade piece you were proudest of has never once surfaced, while a two-year-old byline you forgot about gets cited constantly.
5οΈβ£ Rerun the same prompt set next quarter. One measurement is an anecdote. Two is a trend line, and a trend line you generated yourself beats anybody's survey.
Suggested Tool: Claude (it will tell you your sample is too small rather than building you a confident-sounding case on 18 data points)
Why You'll Love It
- You walk into the budget meeting with data from your own company, the only kind that survives a follow-up question.
- It finds the gap between the coverage you're proud of and the coverage AI actually reads, and those two lists are never the same.
- It gives you the counterargument before someone else does, so you're answering it instead of absorbing it.
The industry will spend the next year arguing about which AI visibility study is right. You don't need to referee that fight. You need forty minutes and your own numbers.
Talk soon,
Chris
β‘ The Cowork Version
Same defense, except Claude runs the prompt set itself, builds the citation log from scratch, and leaves the spreadsheet and the one-pager in your folder.
Setup (one time):
- Make a folder and drop in your coverage report or media monitoring export from the last 12 months. CSV, XLSX, PDF and Word all work.
- Add a short text file with your company name, your three closest competitors, your category in plain language, and your current AI visibility spend by line item.
- Optional: add last quarter's citation log so the comparison runs automatically.
- Optional: connect Google Drive or SharePoint if the coverage reports already live there.
Run this Cowork command:
Build an Earned Media Budget Defense from the files in this folder and save the outputs here.
β
1οΈβ£ Read the context file and write 20 buyer-stage prompts a real customer would ask an AI assistant when comparing us against the named competitors. Show me the list before you run it.
β
2οΈβ£ Run every prompt and capture each source cited: URL, domain, page type, and whether it is owned, earned, or neither. Build this into a structured citation log.
β
3οΈβ£ Read the coverage export and cross-reference it against the citation log. Identify which specific placements appear in AI answers and which never surface. Look for patterns in outlet, recency, format, and whether the piece names us in a headline.
β
4οΈβ£ Calculate the owned, earned and third-party split with raw counts alongside percentages. State plainly what this sample size can and cannot support, and what a defensible sample would require.
β
5οΈβ£ Draft a budget recommendation with a specific allocation, the reasoning per line, and the strongest counterargument stated fairly and then answered. Use only the data in this folder. Do not import industry benchmarks.
β
6οΈβ£ Save the citation log as AI Citation Log.xlsx with one row per citation, and save Earned Media Budget Defense.md as the one-page recommendation.
What you'll get:
- An
AI Citation Log.xlsx showing every source AI cited about you, typed and tagged
- An
Earned Media Budget Defense.md one-pager with a specific allocation and the reasoning behind it
- A list of which placements are working in AI answers and which ones never surfaced
- An honest read on whether your sample is big enough to defend, written before someone in the meeting asks
- A repeatable baseline, so next quarter is a comparison instead of a fresh argument
Want it on autopilot? Ask Claude to save this as a skill and schedule it quarterly. Four runs in, you have a trend line nobody in your category can argue with, because it is the only one built on your own data.
Here are a few articles to help keep you up to speed:
1. Earned Media is PESO's Biggest Playerβ
O'Dwyer's
September 4, 2026
D S Simon Media surveyed 168 PR executives for its "Earned Media in the AI Era" report. 44% ranked earned first for AI discoverability, 99% said increasing earned campaigns matters, and 52% now put more than half their AI discoverability budget behind it. The number worth sitting with: 87% of agency respondents feel they're succeeding at optimizing earned media for AI, against 59% of brands. That 28-point gap is either a value story or a translation failure, depending on which side of the table you're sitting on.
2. Product Pages Get 24% Of AI Citations, Reddit & YouTube 4%β
Search Engine Journal
September 3, 2026
Greg Jarboe pressure-tested a Ten Speed study of 7,387 citation appearances before publishing it, and the piece is worth reading as much for the methodology fight as the finding. Brand-owned content took 88.3% of citations at the evaluation stage, product pages 24.1%, Reddit and YouTube together 4.2%. Jarboe got a mislabeled chart corrected and got Ten Speed to admit the headline split was descriptive only, never statistically tested. Read this next to story one, because the two conclusions cannot both be the whole truth.
3. Meta debuts its Muse AI agent. Will consumers trust it?β
TechCrunch
September 8, 2026
Twelve days after an $18 billion multistate settlement, Meta launched a personal agent asking for access to your email, calendar, payments, health apps and smart home. Sarah Perez lays out the whole trust ledger: the 2011 FTC settlement, the 2019 $5 billion penalty, the 2023 charges, Cambridge Analytica. Watch this one closely. It is a live experiment in whether thorough security documentation can outrun a reputation, which is the exact bet several of your clients are quietly making right now.
4. Apple launches a new way to prove a photo was shot with an iPhone (not generated by AI)β
Nieman Lab
September 9, 2026
Apple Reference Image ships with the iPhone 18 Pro and embeds a signed ID in a photo's metadata proving it came off a real camera sensor. Google's SynthID marks images as fake. Apple's approach says "this is real," and Apple pointedly skipped the C2PA standard that Nikon, Canon and Google Pixel already support. Two competing proof systems with no interoperability, and every visual asset your team publishes is about to need a paper trail under one of them.
5. Two years ago, Meta killed CrowdTangle. Can a new AI tool fill the void?β
Nieman Lab
September 9, 2026
Arbiter, from the nonprofit SimPPL, runs eight to ten models across Facebook, Instagram, X, YouTube, TikTok, Reddit, Bluesky and 4chan to cluster narratives and map when each one first emerges versus when it peaks. Founder Swapneel Mehta takes a direct shot at our category: social listening tools "send you alerts," but by the time the alert fires, "the harms are done." Journalists at DW, Rappler and Chequeado now see the story forming before your monitoring dashboard does.
β