How long this takes to work

Four to six months is the figure we publish, and it is the figure for a company starting from little or no coverage. The first two months are measurement and building, and nothing visible happens in them. Movement, when it comes, shows up first on specific questions with few suppliers competing for them, and last on the broad category questions everyone wants.

The reason is mechanical rather than mysterious. New pages have to be crawled and indexed before anything can draw on them. Corrections to directories and listings move on other people's schedules. Independent coverage takes as long as it takes to earn.

Anyone offering a result inside a quarter is describing something they do not control.

Why four to six and not two

Because three separate clocks have to run, and only one of them is yours.

The first is production: the pages, the corrected descriptions, the structured data. That clock you control, and it is the fastest.

The second is indexing. Nothing can be drawn from a page a system has not processed, and processing happens when it happens. The third is corroboration — a trade body updating an entry, an association listing being corrected, a technical publication running something. Those depend on other organisations, and no amount of urgency on your side speeds them up.

You can compress the writing. You cannot compress the waiting.

What happens in each phase

Months one and two. The baseline measurement, the question set, the map of what exists and what is missing, and the first pages. Nothing changes in the answers, and the deliverable is the record of where you are starting — which matters later, because without it any movement is anecdote.

Months three and four. The first pages have been live long enough to be processed, and the earliest corrections have landed. This is where narrow questions start to move. The broad ones do not.

Months five and six. Coverage has accumulated and the independent sources begin to matter. If the work is landing, this is where it becomes visible across a larger share of the question set rather than in isolated answers.

Beyond six. The work compounds rather than repeats. Pages built in month two are still being drawn from in month twelve, and the question set keeps getting re-run so you can see whether that remains true.

What moves first, and what moves last

Specific before general, always.

"Which supplier handles case packing for glass jars on a low-volume line" has few companies competing to be the answer, and a page that genuinely answers it can become the obvious thing to draw from. "Packaging machinery manufacturers" has hundreds of companies, decades of published material and several multinationals behind it. The first can move in a quarter. The second may never move, and chasing it is usually a mistake.

Entity accuracy is the exception to all of this. If AI systems describe your company using a product line you discontinued, correcting the sources that say so can change the answers faster than anything else on the list — sometimes within weeks, because you are replacing wrong information rather than competing for attention.

What makes it slower

Four things, in rough order of impact.

Starting from nothing. A company AI systems have never described takes longer than one described occasionally but inaccurately.

Several markets or languages. Each language is a separate set of questions and a separate body of evidence, not a translation of the first.

A brochure site. If there is no technical substance to build from, the writing starts earlier and the indexing clock starts later.

A crowded category. Where a dozen established suppliers are already well described, you are competing for a place in the answer rather than filling a gap in it.

What "working" actually means here

A change in the answers to a fixed set of questions, measured the same way each time.

That set has to stay unchanged for the comparison to mean anything — reword a question in month four and you are measuring your own editing. We treat it as a named thing for that reason, and it is defined on Track. The reading it produces is AI Visibility, how often you appear at all across the questions your buyers ask, defined on Baseline.

Traffic is not the measure. Rankings are not the measure. Both answer a different question, and both can move while this one does not — an argument we set out in full in is SEO dead in 2026.

Why we will not guarantee a date

Because we control perhaps half of the inputs, and guaranteeing an outcome you half-control is a promise made with someone else's money.

What we do guarantee is delivery: the Entity Map and Query Universe within thirty days of kick-off, or the engagement ends and that month is refunded. Beyond that we guarantee the method, the measurement and the reporting — and anyone guaranteeing citations, rankings or leads in this discipline is either misinformed or hoping you are.

What this article cannot tell you

It cannot give you your own timeline. Four to six months is what the method is built around; your category, your starting position and how much already exists will move it in either direction, and the baseline measurement is what tells you which.

It cannot promise movement at all. A category may be crowded enough, or a starting position poor enough, that six months of work produces a better-described company and very little change in the answers. That is a real outcome and we would rather say so here than discover it with you in month five.

And AI systems are probabilistic and change continuously. A quiet month is not evidence of failure, and a good week is not evidence of success. Only the trend across a fixed question set is worth acting on.

What to do next

Start the clock with a measurement rather than with production. Building pages before you know which questions you are absent from is how six months becomes twelve.

If you want to run the first reading yourself, we have set out exactly how in can you do this yourself. If you would rather have it done properly — agreed questions, all six systems, your named competitors alongside you — the AI Visibility Audit takes three weeks, and it is the point from which the four to six months is counted.