For twenty years a placements list was a trophy cabinet. Coverage in a Tier 1 outlet, a niche trade, a respected newsletter. You logged it, reported reach, and moved on to the next pitch.
That list now does a second job nobody assigned it. When an AI assistant answers a question about your category, it leans hardest on third-party sources: earned coverage, analyst writeups, independent reviews. Not your homepage. The asset your comms team spent two decades building is the substrate AI engines run on, and in most companies nobody has told them.
The second job your placements list took on
A generative answer can only name a handful of vendors, so something upstream of the writing has to decide who earned the mention. That decision runs on corroboration: what the rest of the web, independently and consistently, keeps saying about each brand. Your own site states a claim. A placement verifies it. Forty independent sources describing you the same specific way is a pattern a model treats as safe to repeat, and the signals engines lift into answers are exactly the kind a strong placement carries: a named customer, a number, a date, a capability stated plainly.
So a placement is no longer only reach. It is evidence sitting where the models read. That changes what the list is worth, and it changes how it should be scored.
Engines do not read the same cabinet
The instinct is to treat coverage as one fact: we got covered. In AI answers it is four separate facts, one per engine.
Each engine reads its own source surface. Grok is web-first and weights live X conversation heavily, with its own taxonomy of source surfaces it trusts. Gemini inherits Google's index and the trust signals that come with it. ChatGPT reads through its own retrieval stack, which favors a different set of outlets again. The niche trade that quietly powers your inclusion in one engine can be completely invisible to another.
This is why a placements list sorted by outlet tier and estimated reach tells you almost nothing about AI visibility. Tier 1 by reach is not Tier 1 by engine trust. A mid-sized trade the models actually cite is worth more, in this channel, than a prestige logo they never read.
A citation this week can vanish next week
Answers are rebuilt on every ask, and the sources behind them rotate. A placement quoted in Monday's answer can drop out of the same answer by Friday. We have written about this churn before: rankings drift, and coarse, stable positions beat precise, fragile ones. The same mechanics apply to your coverage. One screenshot of one answer is a vibe, not a read.
And because most of this influence lands in answers nobody clicks, the churn never shows up in your analytics. A placement can gain or lose its seat in the answer set without a single referral visit changing. If you only measure clicks, you will never see this channel move.
The re-read: score coverage by engine, not by outlet tier
The placements list does not need to be rebuilt. It needs to be re-read with a different question: not where were we covered, but which engine actually pulls from each placement, and does it hold on the re-ask.
- Start from buying questions, not the outlet list. Take the questions your buyers actually ask when they shortlist vendors in your category.
- Run them across the four engines. ChatGPT, Gemini, Claude, and Grok each get the same questions. One engine is a spot check; four is a read.
- Log which placements each engine cites or paraphrases. Placement level, not domain level. A single article can be load-bearing while the rest of the outlet is dead weight.
- Re-ask across a rolling window. Score stability, not one-time presence. A placement cited across weeks is infrastructure. A placement cited once is a coin flip that landed your way.
- Re-sort the cabinet. Cited and stable is working infrastructure. Cited once is fragile. Never cited is inventory: real work, real reach, zero pull in this channel.
Same list. Different question. A completely different picture of what your earned media is worth.
Measured against what? The control
A citation read in isolation flatters. The control that makes it honest is the rival who currently owns your category's answer. Run the same re-read on their coverage: which placements power their inclusion, which outlets every engine seems to trust when it talks about your market. That list of engine-trusted outlets is not trivia. It is next quarter's pitch list, ranked by the only editor that matters now.
This is also where the re-read stops being a reporting exercise and starts being direction. The gap between the coverage the engines cite for the answer owner and the coverage they cite for you is the earned-media work worth doing next.
PR did not get demoted. It became infrastructure.
Nothing about the craft changed. Media relationships, the instinct for a story, the judgment about which placement matters: all of it still decides whether the coverage exists at all. What changed is the scoreboard. Infrastructure gets measured per engine, over time, against a control, and that measurement is exactly the read Trendscoded runs: which sources each engine lifts for your market's buying questions, how your placements hold week over week, and where the answer owner's coverage beats yours.
The question for comms teams is simple, and most have never been asked it: has anyone re-scored your coverage by which AI engines actually cite it?
