Ongoing Legibility
Legibility work is maintenance, not a project. This is what running it continuously looks like.
Measuring how AI systems describe a company became cheap and ordinary in about a year. Fixing what the measurement finds did not, and keeping it fixed is a different problem again.
A single engagement can move a company's position in AI answers. It cannot hold it there. Models retrain on a schedule nobody outside them controls. Competitors publish. Sources get taken down, rewritten or de-indexed. An entity record with no one maintaining it drifts out of date the same way a company registration does. The right comparison is not a website redesign, which ends. It is security, which does not.
So this is the version of the work that runs on a clock: the same protocol, the same frozen question set, re-measured every quarter, with the work in between and a written record of what changed and why.
Baseline Measure before anything is touched
A fixed set of questions, written down and never edited afterwards: identity, category and recommendation. Run repeatedly against named model versions, with languages and markets scored separately. Raw answers kept, not summaries. Everything after this is measured against it.
Structure Decide what is canonical
For a single person this is straightforward. For a company with several brands, a distribution arm and a founder who sells on her own name, it is the whole job: which entity carries the authority, which ones inherit it, which are deliberately kept apart. Get the ordering wrong and a quarter goes into reinforcing the weakest node.
Deployment Make the decision readable
One definition repeated identically everywhere. Structured data on the pages that matter. Public entity records with name variants bridged. Independent sources that describe you the way your own pages do, because a claim only on your own site stays a claim.
Re-measurement Same questions, same method, twenty-one days later
Reported as a share of runs rather than a screenshot, with a log of everything that changed in the period, including the things neither of us controlled. This is the part that separates a result from a coincidence.
Quarter Repeat, and watch for drift
Every quarter the same question set runs again. Some quarters show movement, some show decay, some show noise. The value is knowing which, and acting on the first two rather than the third.
Who this actually suits
- Companies with more than one entity: a retail brand plus a distribution arm plus a founder who sells on her own name. The ordering question is what makes the work worth paying for.
- Businesses in categories where a recommendation is the purchase: what a model says decides who gets considered at all.
- Anyone operating in more than one language or market, where a gain on one side says nothing about the other.
- Not a good fit for a single person with a single site and a single market. Run the Sprint yourself, or take the one-off Audit. Both are the honest answer at that size.
What is not promised
- No guaranteed positions in AI answers. Nobody can offer that, and the offer itself is the warning sign.
- No claim that every quarter improves. Model answers vary across identical prompts, so some quarters will read as noise, and the report will say so rather than dress it up.
- Results stay qualified by the markets, languages and questions actually tested. A gain in one language is evidence about that language.
- Structure amplifies substance and does not replace it. If nothing verifiable exists to find, this work makes the absence easier to read.
Questions
Why would this need to be ongoing?
Because the thing being measured moves on its own. Models retrain, competitors publish, sources go stale and get de-indexed, and an entity record nobody maintains drifts out of date. A structure that made a company legible in one quarter is quietly weaker two quarters later with nobody having touched it. That is a property of the environment, not a sales argument.
How is this different from the Audit or the Diagnostic?
Those are single measurements with a plan attached. This is the same measurement repeated on a schedule, with the work in between and a record of what changed and why. If you only ever want to know where you stand once, take the Audit and stop there.
What does a quarter actually contain?
A re-run of the frozen question set across the same named models, the same number of runs per question, languages and markets kept separate. Then a delta report against the previous quarter, a list of what changed in the environment, and the next set of interventions in priority order.
Can you promise the numbers keep improving?
No, and be suspicious of anyone who does. Model answers vary across identical prompts, so some quarters will show noise rather than movement. What the work guarantees is that you will know which one you are looking at, because the method is fixed and published.
How long is a sensible commitment?
Two quarters minimum, because one re-measurement cannot separate your intervention from the environment moving on its own. Below that, take the one-off Audit instead; it is the honest product for that budget.
What if we have several brands or companies?
Then the ordering question comes first: which entity carries the authority and which ones inherit it. That decision is worth more than any single quarter of maintenance, and it is the case where this work pays for itself fastest.
Start with a measurement
Every engagement starts with the baseline, and the baseline is worth having even if nothing follows it. Tell me what you are working with: how many connected entities, which markets, which languages. I read every enquiry myself and reply within one business day.