Discoverability Baseline

Your buyer already has a shortlist. You are on it, or you are not.

Before anyone calls you, a machine has already narrowed the field. This tells you where you landed, question by question, across the four assistants your buyers actually use. Then it tells you what to change.

$3,000 Your first one is on me

  • Up to 50buyer prompts
  • 4assistants
  • 3competitors
  • 25sources reviewed
  • 12ranked actions
  • 90day sequence
A buyer, this morning

Who are the best outsourcing partners for a company our size?

The answer it gave
  1. A firm that published a cost guide
  2. A firm that published an implementation timeline
  3. A firm quoted in three industry roundups
  4. A firm with named client results in plain text
  5. A firm half your size with twice your published proof

Your name did not come up. Not once.

This happens whether or not you are in the room.

No form was filled in. No call was booked. The field was narrowed to five, and the buyer moved on.

In brief

A fixed panel, scored, and a queue of work against it.

I build the questions your buyers actually ask, in their words, for the segments you sell into. The panel runs against all four assistants and every answer is scored. The questions never move, so the number means something from one run to the next.

01 Where you stand A score per assistant, never blended, because an average hides the one that is failing you. One question names you directly as a control, which shows whether an assistant knows you exist at all. That is a more basic problem than losing a comparison.
02 Who gets recommended instead Every company the assistants put forward, counted, with your own line tracked run over run so you can watch the gap close.
03 What the answers were built from The pages the assistants read before answering. That is your target list. A mention on a page already shaping these answers moves more than a page on your own site that nothing reads.
04 What to do about it A ranked action queue, ordered by what moves the score most for the least work, written against your findings rather than a checklist.
The measurement is the instrument. The work is what moves the number, and the work is what you are buying.
What to expect, honestly

Answer engines take roughly thirty to forty-five days to recrawl a changed site and start citing it, and a URL change usually dips before it lifts. Real movement shows up in the back half of an engagement. Anyone promising a faster curve is guessing.

01

The machine is not judging your work. It is reading what you published.

Delivery record does not enter the answer. An assistant assembles a shortlist from text it can find, parse, and attribute. Firms win these answers by writing the thing the buyer asked about. Most of your proof lives in decks, PDFs and conversations, where nothing can read it.

What it readsGuides that answer the buyer's question directly
What it readsNamed client outcomes written as sentences
What it readsThird-party roundups and directory listings
What it cannot readYour case studies, if the numbers sit inside an image or a PDF
02

I put the questions your buyers ask to the assistants they ask.

Up to fifty buyer questions, written cold from a read of your site and locked before anything runs. Each one goes to ChatGPT, Claude, Perplexity and Gemini. I record whether you were named, whether you were cited, and who took the answer instead.

The question, as a buyer would type it GPTClaudePplxGemini
Who handles this at our size?
What does this cost, realistically?
Who is good in our vertical?
How long does it take to stand up?
Who should we call first?
What is [your company] known for?

Six of the panel, illustrative. Filled circles are answers where the company was named and cited. The last row is the one nearly everyone wins: the question a buyer only asks once they already have your name.

03

What lands in your inbox, and how it gets there.

Eight steps. Every finding traceable back to a query you can re-run yourself.

  1. Scope lock We confirm your category, your ICP, the services that matter commercially, the markets you sell into, and which assistants we test.
  2. The panel — up to 50 prompts I write up to fifty real buyer questions tailored to your offer, so the test measures what actually carries revenue rather than what is easy to score.
  3. What the models say about you today A record of how each assistant describes you right now: who it thinks you are best for, what it gets right, and what it leaves out entirely.
  4. Three competitors, the same questions The identical panel run against three competitors you approve, so you can see who gets recommended instead of you and what evidence earns it.
  5. The source log — top 25 The twenty-five sources the assistants lean on most across your panel, and the ones where your name is missing from the record.
  6. What they get wrong Every incorrect, outdated or ambiguous statement the models make about you, with what to change so the risk goes away.
  7. Marching orders — 12 ranked actions Twelve actions in priority order, starting with whatever is most likely to move accuracy and get you recommended.
  8. The first ninety days A practical sequence for those twelve: what to do first, what comes next, and what can wait.
04

Findings are easy. The value is in the actions.

Twelve changes, ranked, each with an owner and the queries it moves. The ranking comes out of twenty years inside this industry: BPO media relations, analyst relations, building customer review programmes, and the marketing that ties all three together. That experience is what decides which action goes first, and which one reads well in a deck while moving nothing. Five of the twelve here, to show you the shape of it.

First Publish the cost and timeline guide your buyers keep asking for, with real ranges from work you have already delivered. Content · moves 2 questions
First Move your client results out of the deck and onto a page, as plain sentences with the numbers in the text. Website · moves 3 questions
Next Build the page for the vertical you actually win in, structured like the pages currently taking that answer. Website · moves 1 question
Next Get listed where these answers are assembled from, starting with the four directories that fed your competitors. Reviews · moves 3 questions
Later Re-run the identical panel and find out which of the above the machines actually noticed. Pivot AI Global
05

Then we find out whether it worked.

One Baseline is a photograph. The panel is fixed and immutable, so running it again on the same questions is the only honest way to know whether the work moved anything. That is the engagement: ninety days, the panel re-run weekly, the work steered by what moves. That part is the paid part, priced to your panel.

Baseline Day 30 Day 60 Day 90 Q5
Forbes · Aug 2026 The New Brand Test: Can AI Describe Your Company Accurately? — Andrew Kokes, for the Communications Council Read the article ↗

Also in Forbes, four days later: 16 Smart Ways To Use AI To Make A Company Blog More Engaging — expert panel contributor, Aug 25.

Start here

Just two fields to get started.

I run the first Baseline for a company at no charge. You get the whole report, not a sample of one.

  1. You give me your website and a work email.
  2. I read the site and come back with the segment you compete in, the buyers who would be searching, and the three competitors you are really measured against.
  3. You approve that scope, or tell me where it is wrong and I rebuild it from your answer.
  4. The report lands. No card, no invoice.

One boundary, stated up front. This sits on the commercial side of the business. It is separate from the IPO 25 and the Pivot Path rankings. Requesting it, paying for what follows, or declining to, has no effect on your position on anything we publish.