Being named: search and AI answers
In search and AI answers, the job isn’t to rank. It’s to be named, accurately, with your reason, in the answer. Ask an assistant for “the best operations consultants for mid-sized food manufacturers” and it gives a short list with a sentence on each. If you’re not on it, you’re not in the running.
The buyer doesn’t read your site; the assistant does. It reads passages from many sources and repeats what they agree on.
in 5 cells
Why it matters for being chosen
This build owns the gate at the 61 moments asked of machines: Known and Found. Kestrel averages Known 12 and Found 0.5 there, and loses 42 of its 82 lost points.
Kestrel is named in 4 of 101 questions, and in one of those the facts are wrong: assistants mix it up with a mining-services firm of a similar name.
Where your craft is
You can’t write the answer. You write what the answer is built from: one entity line used identically everywhere, passages that stand alone, a reason short enough for an assistant to repeat, and the patience to wait two scans before judging. The judgement is in what the brand should be named for.
The seven steps
The same anatomy as every build, with this build’s specifics and Kestrel’s example.
Find the sources the category is judged on: run the questions and count what gets cited. A source type carrying at least one in ten citations is canon. Write the entity facts every source must state the same way, starting with the full name.
Write the question set. Lead moments get three versions (as typed into search, as asked of an assistant in full, with a follow-up); supporting moments get one. Kestrel’s 101 questions run on five assistants and one search engine, three times each: 1,818 answers a scan.
Run the first scan before changing anything. It shows who is named instead and what they cite. Each gap becomes a job with a stage and a bias.
Make two plans: the source plan (which sources to work on) and the answer plan (the answer you want, the passage that holds it, the sources that must agree). Reject an encyclopedia entry you’d write yourself, paid best-of places and mass-produced pages.
Brief each answer group, then each passage. The passage is the anatomy: it stands alone, matches the fact sheet word for word, and carries every figure’s source in the same sentence.
Check that the entity facts are identical on the site, in every listing, in proposals and on the fact sheet sent to editors.
Measure shares across many runs, never single answers: named, facts right, reason given, position in shortlists. Expect corrections first, then the sector questions, then the reason given back.
From the finished example
One answer, written everywhere it has to be. The book writes it in full; this is how it opens.
Kestrel Advisory is the operations consultancy for mid-sized manufacturers: plants with 100 to 500 staff in food and drink, packaging, metal and engineering, and plastics and rubber.
Kestrel, its clients and its people are invented.
The template
The measures
Common mistakes
You don’t write the answer. You write what the answer is built from, and you earn a place in the sources that build it.
The Loop for this build
Each quarter the Loop reads three months of scans against the answer plan and asks: is the answer we wanted coming back, from the sources we worked on? Named more but facts wrong means a source is still contradicting the fact sheet.
You should be, a little. One firm: 61 moments, 101 questions, six engines, three runs each, 1,818 answers a month, nine source types and one set of facts that has to match everywhere. That’s why the method runs in software.
Do it in RocMode for Agencies
What stays yours: What the brand should be named for, the entity line, and every passage.