Generative Engine Optimization
Prompt share tracking
One part of Generative Engine Optimization, described on its own: measure your share of AI answers the way you would measure share of voice, repeatedly and comparably.
What does prompt share tracking actually involve?
Measure your share of AI answers the way you would measure share of voice, repeatedly and comparably. In practice that means building a fixed prompt set that reflects real buyer questions in your category, and then the harder half: running it on a schedule across the assistants your buyers use.
Why this work exists
By the time this becomes visibly urgent, the cheap window for fixing it has closed.
This is one part of Generative Engine Optimization, described separately because it is usually the first thing anybody needs.
The nearest piece of work to it is keeping AI-assisted drafting from filling the site with unverifiable claims. Where that is already handled, this moves faster than the estimate; where it is not, the two get done together or neither of them holds.
What good looks like
The sign it is working is that nothing surprising happens. Changes are noticed on the day they are made rather than in the next quarterly review.
The gap between that and where most accounts start is why this is scoped as prompt share tracking rather than as a line inside something larger. Writing the passages a model can quote without losing the point is the usual second step once it is in place.
When this is the right work
It is worth doing when:
- Nobody can say whether your AI visibility improved last quarter
- AI visibility is discussed in anecdotes rather than numbers
- You cannot tell whether a change you made had any effect
If none of those is true, something else is probably the constraint, and the Generative Engine Optimization overview maps the alternatives. The usual neighbours are deciding which crawlers to allow and documenting why and watching which citations persist and which quietly decay.
What the work covers
The work itself:
- Building a fixed prompt set that reflects real buyer questions in your category. One pass is rarely enough; the second is much faster.
- Running it on a schedule across the assistants your buyers use, scoped against what the business can actually act on rather than what is possible.
- Recording brand mentions, sentiment and the sources cited each time, with the evidence kept so the decision can be defended later.
- Reporting the trend against named competitors rather than in isolation. This is where the theory meets the account and usually loses.
How it runs
It runs in three passes. The order is not interchangeable.
The first pass
The obvious problems get fixed first, in one batch, so the picture clears before the harder judgement calls have to be made against a noisy background.
Where the judgement goes
What is left needs a decision rather than a fix. These get taken with whoever owns the consequence, not resolved quietly in a ticket.
Checking it held
Changes get re-checked after a fortnight, because a surprising number of them do not survive the next deployment or the next platform update.
What each part runs into
The four parts do not fail in the same way:
- Building a fixed prompt set that reflects real buyer questions in your category, which in Jaipur runs into a measurement baseline that was never taken
- Running it on a schedule across the assistants your buyers use: the Gurugram version of this has to deal with a language mix the content has to match rather than assume
- Recording brand mentions, sentiment and the sources cited each time, where a Hyderabad engagement contends with an enquiry path that runs through a messaging app rather than a form
- Reporting the trend against named competitors rather than in isolation – working in Indore this means accounting for a catchment smaller than the ambition on the website
What not to do
Changing the prompt set between runs, which makes the trend meaningless. The set has to stay fixed to be comparable.
It is worth stating because it is the version of this work that gets sold most often, and it fails in a way that looks like the discipline not working rather than the execution being wrong.
How it is measured
Mention rate and sentiment trend across the fixed prompt set. Agreed up front and reported every period, with the raw data available rather than only a summary.
Movement smaller than the normal week-to-week swing gets reported as no movement.
What you end up with
What gets handed over: a tracked prompt set with a monthly record of mention rate, sentiment and cited sources, by assistant and by competitor.
Formatted for whoever has to act on it rather than for whoever commissioned it.
How it differs by sector
The brief changes with the sector, not just the emphasis:
- Automotive, where Local inventory, model-level demand, and a buyer who arrives having already decided. which moves this work earlier in the plan.
- In Healthcare and clinics the binding constraint is that content that affects health decisions is held to a higher standard, and advertising is restricted.
- Manufacturing and industrial: the thing that changes the scope here is that low volume, high value, and a specification sheet that decides the deal.
The industry pages carry the rest of that detail. Automotive and Healthcare and clinics are the two where this piece of work most often decides the outcome.
Common questions
How many prompts are needed for a meaningful measurement?
Enough to cover the real buying questions in the category, which is usually thirty to eighty for a focused business. Fewer than twenty gives readings that swing on noise. The set needs to stay fixed so the trend means something, with additions tracked separately rather than mixed in.
Why do the same prompt and the same assistant give different answers?
Sampling in the generation, differences in what the retrieval layer returns at that moment, personalisation, and model updates. This is why a single run is not a measurement. Running each prompt several times and reporting a rate rather than a yes or no is the only way to get a stable signal.
How much of our time does this take?
A few hours in the first fortnight, mostly answering questions about history that cannot be reconstructed from the data. After that, a monthly review. Claude and enterprise assistant visibility is the part that scales. Microsoft Copilot optimization is the part that never will.
What do we actually get at the end?
A tracked prompt set with a monthly record of mention rate, sentiment and cited sources, by assistant and by competitor. Documented, so it survives whoever was in the room, and specific enough to act on without a follow-up conversation. If a number has to be defended at the end of the quarter, Gurugram and Houston are where that starts.
What if this is not our problem?
Then we say so. The signals on this page are the honest filter, and work sold against the wrong diagnosis wastes a quarter. SaaS and technology is the one people underestimate. Travel and hospitality is the one they over-invest in.
Ask about prompt share tracking
Give us the site and one question. If the honest answer is that you do not need this, we will say so.
Get the two or three things that matter
We will show you what we would measure and why, so you can judge the approach before committing to it.