The verdict first, since you will build a forecast around it: technical wins land in weeks, visibility movement typically shows inside the first quarter, and citation share compounds slower than both. A serious AEO program is designed around that physics, not against it. Everything below is the work in the order it should happen, and the order is the part most teams get wrong.
A word on what "serious" means here. It means a versioned measurement system, shipped fixes rather than ticketed recommendations, and a day 90 readout that admits what did not move. If a proposed program is missing any of those three, it is content marketing wearing a lab coat.
Days 1 to 30: baseline and unblock
You cannot claim movement without a day-zero number, so the first job is the baseline. That means measuring visibility, share of voice, and sentiment on a versioned prompt set, run against named competitors. Versioned is the load-bearing word. Prompt sets change as you learn what buyers actually ask, and if you cannot say which version of the set produced which number, your quarter-over-quarter comparison is fiction. We walked through the mechanics in your first citation share baseline.
The second job is the crawler-eye audit: looking at your site the way retrieval systems see it, rather than the way your designer does. What renders empty without JavaScript. Which robots rules quietly turn away AI crawlers. What sits behind a form. Teams are routinely startled by this view, because a site that looks polished in a browser can read as a locked building to the systems assembling answers about you.
Third, entity cleanup. Engines resolve you as an entity: one name, one category, one set of facts. When your site, your directories, your profiles, and your press coverage disagree about what you are called and what you do, the engine's confidence drops and its answers get vague or wrong. Month one is when those records get reconciled everywhere engines look.
By day 30 you should be holding three things:
- A versioned prompt set with visibility, share of voice, and sentiment scored against named competitors
- A crawler-eye audit with the access fixes shipped, not ticketed
- Entity records that agree with each other everywhere engines read
Shipped, not ticketed, is the tell. A diagnostic that ends in a slide deck is a diagnostic that ends.
Days 31 to 60: build what engines can cite
Month two is construction. Take the highest-weight prompts from the baseline, the questions with the most buying intent behind them, and build citation-ready pages for them: pages that answer the question directly, state facts a machine can lift cleanly, and carry structured data that tells the engine exactly what it is reading.
This is also when canonical proof pages go ungated. Engines cannot cite a PDF behind a form, and neither can anyone else. Your strongest evidence needs one authoritative page per story, open to the crawl, with the numbers and the method on the page itself.
Third-party footprint work begins here too, because engines corroborate before they cite. Reviews, directories, press, analyst mentions: the places an engine checks your claims against. This is the slowest workstream in the entire program, which is exactly why it starts in month two instead of month three.
None of this is exotic. One anonymized example: a national legal firm's attorneys page, rebuilt along these lines, became the most-cited page in its space out of more than 1,900 pages we track. Not the homepage. Not a thought-leadership essay. The plain page that directly answered the question people were asking.
Days 61 to 90: expand, measure, decide
Month three pushes coverage down the prompt list, from the head questions into the longer tail of comparisons, pricing queries, and category questions. The pages come faster now because the pattern is established and the technical plumbing is done. Expect the tail to matter differently as well. Head prompts build presence; tail prompts sit closer to shortlists and demos, so coverage down the list is where the pipeline story usually starts.
This is also when the first measurable movement typically appears: citations landing on the pages built in month two, visibility ticking up on specific prompt families, sentiment shifting where the entity work has settled. Typically. Not always, and never evenly across engines, which is why the quarter ends with a readout rather than a highlight reel.
The readout states what moved and what did not, per engine and per prompt family, with no varnish on the misses. Then come the kill-or-scale decisions. Page patterns that earned citations get siblings. Workstreams that produced nothing measurable get killed, in writing, so next quarter's budget flows to what the evidence supports. A readout that cannot name a miss is the readout you should trust least.
What to expect, plainly
Set expectations by workstream, not by mood. Technical wins land in weeks: access fixed, entities consistent, pages readable by the systems that matter. Visibility movement typically shows inside the first quarter, on some engines before others. Citation share compounds slower than both, because it depends on third parties and repeated re-verification, and no amount of enthusiasm accelerates other people's websites. The compounding is real, though. Citations beget citations, since each one adds corroboration for the next engine that checks, which is why quarter two typically outperforms quarter one on the same effort.
So hold your program to the right clocks. Demand the technical fixes fast. Expect visibility signal by the readout. Give citation share the quarters it needs, while checking monthly that it is compounding rather than flatlining. Anyone promising top citations by week three is describing a different business, or no business at all.
The full program, including what it costs, is laid out at AI search optimization, pricing included, because we would rather lose a prospect at the page than in the third meeting.
Ninety days from now you will have either a readout or a pile of impressions, and only one of those supports a decision.