Most AI visibility work fails for one of two reasons: it starts with content before measuring, so nobody knows what to write, or it starts with technical fixes and stops, so nothing is ever written. This playbook orders the work so each month's output feeds the next, with timing that reflects how fast engines actually re-read the web.
Days 1 to 15: Measure and unblock
Build the question set. A few hundred real buyer questions across category, selection, comparison, and fact questions. Fix it; everything is measured against it.
Run the baseline. Every question through every engine. Share of answer per engine, competitor share, source log, and accuracy flags.
Check the plumbing. robots.txt and CDN rules for every AI crawler, JavaScript rendering on key pages, schema validity, llms.txt presence. Fix blocks immediately. Nothing else matters while engines cannot read you.
Decide the targets. From the source log, the five to eight domains engines cite most for your category. From the accuracy flags, every wrong fact and where it lives.
Days 16 to 45: Write and align
Write the three pages. Your head-to-head comparison against the competitor you lose to most, the category shortlist page for your core buyer, and the alternatives page for the category leader. Each answer-ready: verdict first, criteria, table, honest trade-offs, FAQ with schema, dated.
Rewrite the pricing and product pages to answer-ready standard, with facts in the HTML and Product or SoftwareApplication schema that matches.
Run the consistency pass. Canonical fact sheet, Organization schema with sameAs, llms.txt, Wikidata item, every review and directory profile claimed and aligned.
Correct the wrong facts. Your own stale pages first, then profiles, then third-party articles with a polite request.
Timing note: engines re-read your own pages within days to a couple of weeks. Expect the first movement on fact questions and comparison citations toward the end of this window.
Days 46 to 75: Get into the sources
Review platforms. Complete profiles on the two or three engines cite, an honest review-request flow tied to real customer milestones.
Communities. A named, disclosed expert from your team participating in the two or three communities engines cite, answering fully, correcting stale threads factually.
Lists and press. Pitch the publications whose "best of" lists engines cite, with real differentiation and current facts.
Video. Outreach to the reviewer channels engines cite; set up your own channel so each video answers one question with a text description.
This is the slowest-moving work and the longest-lasting. Some of it will not show in share of answer until after day 90. Start it anyway, because it compounds.
Days 76 to 90: Track and adjust
Weekly share of answer on the fixed question set, per engine, with the source log refreshed and what-shipped next to what-moved.
Read the pattern. Which engines moved, which did not, and why. A Perplexity gain with no ChatGPT movement usually means the comparison pages worked but the reputation sources have not caught up. A ChatGPT gain with Gemini static often points at a Google-specific source you are still missing.
Reallocate. Drop what did nothing, double what moved, and set the next quarter's targets from the current source log, which will have shifted.
What realistic progress looks like
Measured in share of answer, not traffic. A company starting from near-zero commonly sees fact questions and comparison citations move first, selection questions ("best X for Y") move second as source presence builds, and training-based answers move last, sometimes outside the 90 days. Gains on the earliest movers usually appear within the first six weeks when the plumbing was the problem, and later when the problem is reputation.
Running it with help
AEO GEO Labs runs this sequence as its standard engagement: visibility audit, pages written to be quoted, the consistency pass, getting into the sources, and a weekly report showing share of answer per engine next to what was shipped. The weekly pairing is the discipline that keeps the work honest, because it makes it obvious when something did not work.
Day one
Build the question set and run twenty of the questions yourself tonight. The baseline will tell you which of the above to start with. If you want the full baseline run live, with every engine and the source log, AEO GEO Labs does it on a 20-minute call.