Earned

Being Named: the Search and AI Answers Playbook

The idea

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. Many buyers take that list as it comes. If Kestrel isn't on it, Kestrel isn't in the running. If it's on the list with the wrong facts, that's worse.

Three things make this build different from the others:

• The buyer doesn't read your site; the assistant does. It reads passages from many sources and repeats what most of them agree on. So half the work is on your own pages, and half is in the sources assistants trust.
• Answers vary. Ask the same question twice and you can get different firms. So measure shares across many runs, never a single answer.
• Traffic misleads. A buyer who sees your name in an answer, then searches for you or calls, may never arrive through a link. Measure being named and being right, not clicks.

The method is the same as every build: the shape (the sources the category is judged on), coverage (the questions from the lit moments), reconcile, brief, check and measure.

The footprint, and Kestrel in 60 seconds

This build leads every question Kestrel's buyers put to an assistant before the pitch: what's wrong, what the options are, what a good review should cover, who does this, and what it costs. It supports the COO's and finance director's early searches, the investor's search for firms, and the reputation check every buyer runs before signing: "Is Kestrel Advisory any good?"

Measuring how assistants see you is diagnosis: it belongs to step five, the Selectability Score. Earning a place in their answers is this build.

Kestrel in 60 seconds. Kestrel Advisory is the invented firm used in every playbook: a 60-person operations consultancy that helps mid-sized manufacturers find waste and lift output. Its main offer is a fixed-price, six-week operations review. Its strategy is in Part one, and its full moment map, every chooser's question at every step, in Where you stand, part one.

• Who chooses: usually a committee. The COO starts it, the finance director approves the spend, the plant manager lives with it and procurement runs the paperwork. The owner, the in-house improvement lead, an investor and AI assistants all have a say.
• The reason to be chosen: "We find the savings on the floor in six weeks, at a fixed price, and the review pays for itself or you don't proceed." (Invented. A real firm would need proof for every part of that sentence.)
• The alternatives: a large consulting firm, a specialist lean consultant, the in-house improvement team, buying new equipment, or doing nothing.

Its slice of the Score

This build owns the gate at the 61 moments asked of machines: Known (do assistants recognise Kestrel, and tie it to this moment?) and Found (is it named when buyers ask without naming it?). It shares Believed and Preferred with PR and reviews, wherever Kestrel is named. The numbers come from the Selectability Score; the scan below is the same one, read for this build.

Half of everything Kestrel loses, it loses at these 61 moments, and nearly all of it at Found. No other build has as much of the Score riding on it. That's also why it's slow: Found moves only when the sources assistants read have changed, and they read on their own schedule. Expect the first movement two or three monthly scans after the work starts, not the week after.

Step 1: the shape, the sources the category is judged on

For a website, the shape is the page types buyers expect. For search and AI answers, it's the sources assistants and search engines draw on when someone asks about the category. Every category has its own mix, and it's knowable: run the questions, count what gets cited. For B2B professional services sold to manufacturers, the mix usually looks like this:

The entity runs through the shape. Before anything is published, write down the facts every source must state the same way:

• The name, with what makes it unique: "Kestrel Advisory, the operations consultancy for mid-sized manufacturers". Never plain "Kestrel", which is shared by other firms.
• Who it's for: manufacturers with 100 to 500 staff, in food and drink, packaging, metal and engineering, and plastics and rubber.
• The offer: a fixed-price, six-week operations review, on site.
• The way out: if the review doesn't find more than it costs, the client doesn't go ahead.
• The proof pattern: a named plant, a before-and-after number, the plant manager's words.

Assistants repeat what most sources agree on. A fact stated five different ways in five places gets dropped, or mixed up with another firm's.

Step 2: coverage, the question set

The coverage unit for this build is each moment's questions, and who gets named today. Start from the footprint: every lit cell, split by problem and sector where the question changes. That's 61 moments: 20 this build leads and 41 it supports.

Lead moments get three versions of the question, because buyers ask the same thing three ways:

1. As typed into search: "operations review cost"

2. As asked of an assistant, in full: "What does an operations review cost for a 200-

• person food plant, and what does it usually return?"

3. With a follow-up or a constraint: "Is fixed price better than day rates for that?"

Supporting moments get one version each: the lead build carries the rest.

101 questions. Each one runs on five assistants and one search engine, three times each, because answers vary. That's 1,818 answers in a single scan, every one recording whether Kestrel is named, who's named instead, what's said about each, and which sources were cited. And because answers change as sources change, the scan runs again every month.

Three rules for writing the questions:

• Use the buyer's words, not yours. They come from the cell: the top question, the pain, the objection. Nobody asks for "operational excellence partners".
• Leave your name out, except in reputation questions. A question that names Kestrel tells you nothing about whether a buyer would find it.
• Keep the set stable. Change the questions and you can't compare one month with the next. Add new ones when the grid changes; retire old ones with a note.

Step 3: who gets named today, then gaps, biases and jobs

Run the question set before changing anything. The first scan gives two things: the sources the category is actually judged on, and where Kestrel stands in each moment. All the results below are invented for the example.

The sources. Count every citation across the 1,818 answers and group them by source type. A source type cited in at least one in ten citations is canon for this category: it's where assistants look, so it's where Kestrel has to be.

Where Kestrel stands. A question counts as "named" if Kestrel appears in at least one engine's answer, in any of the three runs. Kestrel is named in 4 of the 101 questions, and in one of those the facts are wrong.

Each gap becomes a job: a stage, a moment and a bias.

Every job names a moment, a stage and a bias. That's what stops this build turning into "more content".

Step 4: reconcile, the source plan and the answer plan

Put the jobs into the shape. Two plans come out of it: which sources to work on, and what answer each group of questions should give.

The source plan

The same three signals as the website set each source's priority. Canon: the source type carries at least one in ten citations. Coverage: it carries jobs from Step 3. Strategy: the positioning needs it.

Rejected, with reasons:

• An encyclopedia entry. A 60-person firm is unlikely to meet the notability rules, and writing your own entry breaks them. It would be removed, and the attempt would be visible.
• Paid places in best-of lists. A paid placement is an ad. It belongs in the advertising playbook, labelled as one.
• Mass-produced articles. No moment needs them. Thin pages dilute the passages that do the work, and contradict the proof pattern.
• Questions outside the grid (lean certification, factory software reviews). No chooser asks them on the way to buying a review.

The answer plan

For each group of questions, write the answer Kestrel wants an assistant to give, the passage that holds it, and the sources that must agree with it. This is this build's equivalent of a sitemap.

The coverage test: every one of the 61 moments has a target answer, a passage and at least one source. Every canon source type has a plan. And one check the website doesn't need: every source states the entity facts the same way. If one directory says "100 to 500 staff" and another says "mid-market", fix one before writing anything new.

The anatomy library

For a website, the anatomy is the page. For this build it's the passage: the few sentences an assistant lifts to answer one question. Each kind of question wants its own passage shape. These are the eight Kestrel's answer plan uses.

Three rules hold for every passage:

• It stands alone. No "as mentioned above", no pronoun pointing at the previous paragraph. An assistant lifts it without its neighbours.
• Its facts match the fact sheet word for word. "100 to 500 staff", never "mid-market" in one place and "SMEs" in another.
• Every figure carries its source in the same sentence. A figure whose source is two paragraphs away gets quoted without it, or not at all.

Step 5: the answer brief, then a brief for every passage

Assistants don't quote pages. They quote passages: a few sentences that answer one question on their own. So this build briefs at two levels. The answer brief holds what no single passage can: the questions, what's said today, the answer wanted, the facts that must match, and the sources to earn. Then each passage gets its own brief, like a section on a web page.

Here's the answer brief for one group of questions:

ANSWER BRIEF Comparing, food and drink
QUESTIONS (4) "Best operations consultants for mid-sized food
manufacturers?" + two other versions; COO: "Who does this for food plants our
size?"
MOMENT Assistant and COO · comparing · food and drink · stage: Found, then
Preferred · bias: good enough beats best
TODAY Not named. Named: two large firms, a lean specialist. Cited: two
comparison articles, a trade magazine, a large firm's site
THE ANSWER WE WANT Kestrel Advisory, the operations consultancy for mid-
sized manufacturers, runs fixed-price, six-week operations reviews in food
and drink plants, working on the floor on changeovers and hygiene downtime. A
named plant, its before-and-after figure and its plant manager's words
FACTS THAT MUST MATCH Name and descriptor; 100 to 500 staff; the four
sectors; fixed price; six weeks; on site; the way out
PASSAGES Food and drink page: "Who we work with" and "What a review finds in
a food plant"; the food case page; Price and terms: "What it costs"
SOURCES TO EARN The two cited comparison articles (fact sheet + food case);
the food manufacturers' association member list; two reviews from food
clients
GUARD No claim an assistant can't check ("leading", "best"); every figure
sourced and dated; the full name every time

Then each passage gets its own brief. Two of the four:

PASSAGE Food and drink page · "Who we work with"
ANSWERS "Who does operations reviews for mid-sized food manufacturers?"
FIRST SENTENCE The answer, with the entity facts: who, for whom, what, at
what price model
THEN The two problems food plants bring most, in the plant manager's words;
one named plant
LENGTH Under 80 words, so it stands alone when lifted
ENTITIES Kestrel Advisory; operations review; food and drink manufacturing;
the named plant
GUARD Reads correctly with nothing around it; matches the directory sentence
word for word

PASSAGE Price and terms · "What it costs"
ANSWERS "What does an operations review cost, and what does it return?"
FIRST SENTENCE The fixed price range, for which plant sizes
THEN What's included; the typical return across reviews, with its source and
date; the way out, in the contract's own words
LENGTH Under 80 words
ENTITIES Kestrel Advisory; operations review; the price range; the results
page
GUARD The range matches proposals and every listing; the return figure comes
from the results page, never rounded up

The flow check is different here. A web page is read top to bottom; a passage is lifted on its own. So read each passage alone, as an assistant would quote it. If it needs the paragraph before it to make sense, rewrite it.

The finished example: one answer, written everywhere it has to be

One answer from the plan, done in full: the food and drink shortlist. The question is "best operations consultants for mid-sized food manufacturers". Today, assistants name two large firms, a lean specialist and an engineering firm, and never Kestrel. Here's every piece it takes to change that. Kestrel, its clients and its people are invented.

1. The entity line, fixed once and used everywhere:

• 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. Its fixed-price, six-week operations review finds savings on the floor, and if it doesn't find more than it costs, the client doesn't go ahead.

2. The shortlist entry, the first passage on the food and drink page:

• Operations reviews for food and drink manufacturers with 100 to 500 staff. Kestrel Advisory reviews food plants on the floor, over six weeks, at a fixed price. In 2025 it reviewed Calloway Foods, a 310-person sauce and dressings maker: scrap on the main filling line fell from 4.1% to 2.6% within three months, confirmed by Calloway's finance director. If a review doesn't find more than it costs, the client doesn't go ahead.

3. The directory listing, corrected in the four association member lists and the two consultant directories assistants cite:

• Kestrel Advisory · Operations consultancy · Manufacturers with 100 to 500 staff · Food and drink; packaging; metal and engineering; plastics and rubber · Fixed-price, six-week operations reviews · Not affiliated with Kestrel Mining Services.

4. The correction request, sent to the directory whose old entry assistants mix up with a mining-services firm:

• Our listing combines two firms. Kestrel Advisory is an operations consultancy for manufacturers; Kestrel Mining Services is unrelated. Please replace our entry with the attached text, and remove the mining sector and the old address. Our website states the same facts at kestreladvisory.example/about.

5. The fact sheet for editors, one page, sent with a sector case to the three comparison lists that make up a fifth of citations: the entity line; the offer, price range and way out; the four sectors and size band; one case per sector, each confirmed by the client; the two people editors can quote, with their plant experience; and the benchmark's headline figure, with its method.

6. The structured data on the about page, stating only what the page says:

{"@context": "https://schema.org", "@type": "ProfessionalService",
 "name": "Kestrel Advisory",
 "description": "Operations consultancy for manufacturers with 100 to 500
staff",
 "areaServed": "Australia",
 "knowsAbout": ["Food and drink manufacturing", "Packaging manufacturing",
                "Metal and engineering", "Plastics and rubber"],
 "sameAs": ["[association member page]", "[consultant directory page]",
            "[company profile]"]}

7. The answer Kestrel wants back, the test the monthly scan applies:

• "...Kestrel Advisory works with food manufacturers of 100 to 500 staff, on a fixed-price, six-week review; it reports a 37% scrap reduction at Calloway Foods..."

If the assistant names Kestrel, gives the size band, the fixed price and a food case, the answer has landed. If it names Kestrel with mining in the same sentence, step 4 hasn't. Six pieces of work, in six places, for one question in one sector. There are three more sectors, and six more groups of questions in the answer plan.

Templates

The entity fact sheet (one per firm; every source copies from it)

FULL NAME [the name, plus what makes it unique]
NOT TO BE CONFUSED WITH [firms with similar names, and how they differ]
ENTITY LINE [one sentence: what it is, who for, the offer]
WHO FOR [size band, sectors, regions, word for word]
THE OFFER [what, how long, how priced]
THE WAY OUT [the guarantee or exit, as in the contract]
PRICE [the range, and what sets it]
PROOF PATTERN [what every case states]
PEOPLE TO QUOTE [names, roles, experience]
LINKS [site, profiles, listings: every place the facts appear]
LAST CHECKED [date, and by whom]

The question set (one row per question; never edited, only retired)

The answer brief (one per group of questions)

QUESTIONS [the IDs in this group]
TODAY [who's named, what's said, what's cited]
THE ANSWER WE WANT [in one or two sentences]
FACTS THAT MUST MATCH [from the fact sheet]
PASSAGES [type from the anatomy library, and where each lives]
SOURCES TO EARN [from the source plan, with what each needs]
PRIORITY [P1-P3, and why]
DONE WHEN [what the scan must show, for how many months running]

The passage brief

PASSAGE TYPE [from the anatomy library]
THE QUESTION [answered in the first sentence]
JOB [stage · chooser and moment · bias]
FACTS [from the fact sheet, word for word]
FIGURE + SOURCE [in the same sentence]
LENGTH [from the anatomy library]
GUARD [passes only if it reads correctly on its own]

The scan log (one row per question per month)

Step 6: the business check

• Revenue: the sector and cost questions sit closest to new work. If the plan spends most effort on framing questions, it's chasing attention, not choices.
• One truth: the entity facts are identical on the site, in every listing, in proposals and in the fact sheet sent to editors.
• The proof is real: the benchmark report needs enough reviews behind it, and clients' permission to use their data. Without both, it isn't published, and the trade press plan waits.
• Reviews are earned: asked for after the results are in, from clients who agree, never paid for or written for them.
• Capacity: someone owns the monthly scan, the listings and the press relationships. If nobody does, start with the four P1 jobs: the sector sentence everywhere, the price passage, the mix-up fixed, and reviews.

Step 7: the measures

Measure shares across many runs, never single answers. Report by group of questions, by engine and by month.

From the first scan, Kestrel is named in 4 of 101 questions, with wrong facts in one. A plan that works shows up in that order: facts corrected first, then named in the sector questions, then the reason given back. Clicks barely appear in this list. Most of the value happens before anyone clicks.

In the campaign plan

This build rarely leads a campaign, but every campaign leaves something behind for it: a page, an article, a review, a mention in a source assistants read. The rule in the campaign plan: every campaign names the questions it should move, and the scan checks them a month or two later.

A campaign that leaves nothing for an assistant to read has reached only the people who saw it that week.

Running it

Who owns it. At Kestrel, the marketing lead owns the question set, the fact sheet and the monthly scan. PR owns the press, list and review relationships. The managing partner signs the fact sheet, because every source copies it.

Correcting an assistant. You can't edit an answer. You can fix what it reads. Find the sources it cites for the wrong fact, correct those you control, ask the owners of the rest, and make sure your own pages state the right fact in the first sentence. Then wait two scans before judging.

What not to do. No paid places presented as editorial, no reviews written for clients, no thin pages written to catch questions. Assistants and search engines are built to discount them, and the first two would break the claims rule: nothing said that Credentials can't sign.

Common mistakes

The bias to watch in the team: the illusion of control. 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 (step 11) reads three months of scans against the answer plan, one group of questions at a time, and asks: is the answer we wanted coming back, from the sources we worked on?

What Kestrel's first year would look for. Corrections first: the mining mix-up gone within two or three scans of the listings being fixed. Then the sector questions, as the sector pages and directories line up: named in some food and packaging answers by mid-year. The framing and options questions last, because they depend on the benchmark being covered, and that comes once a year. If Kestrel is named in a quarter of the comparing questions by the year's end, with its reason, the build is working, even though the Score will still say "rarely selected" at those moments. Found moves first; the Score follows.

Shapes by business type

The source mix changes with the business. Run the questions and count the citations to find yours; these are the usual starting points.

Feeling a bit overwhelmed?

You should be, a little. That was 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. Most teams try a handful of questions in one assistant and call it done. The answers change every month, so a one-off check goes out of date fast.

RocMode does the heavy part for you:

• Your questions, built from your buyers. RocMode maps your choosers and their moments, then writes the questions they ask, in their words.
• Run across leading AI assistants. Each question runs on five assistants, three times each. It records whether you're named, who's named instead, and which sources the assistants cite.
• Scored, with the biggest lever named. Your selectability, moment by moment and stage by stage, with a range on the score, questions that name you kept out of Found, and the fix worth the most points.

Join the RocMode beta