The First 90 Days

This chapter is the whole book compressed into a runnable sequence: ninety days from no program to a working one, with exit criteria at each stage and a link to the chapter that explains each step. It also works as a re-entry index: come back with a specific job to do, find the week it belongs to, and follow the link.

Set expectations before day one. Ninety days is enough to fix access, correct facts, build measurement, and see retrieval-side movement, because retrieval reads current pages and current rankings. It is not enough to change what models remember about you from training, so memory-side share moves on the timescale of model releases. Promise the first; schedule the second.

Weeks 1-2: baseline and audit

Build the prompt inventory first, per Prompt and Topic Research: the prompts that precede purchases in your category, weighted by commercial value, with personas where your category is persona-sensitive. Then take the baseline, per Measuring AI Search Visibility: through the API, run each core prompt ten times with search off and ten times with search on, and record share of answers for you and each competitor in both modes. Run the diagnostic from Auditing and Diagnosing Visibility Problems and benchmark the incumbents per Competitive Analysis.

Exit criteria:

• A written prompt inventory
• A baseline spreadsheet with memory-mode and searched-mode share for you and your top competitors
• A diagnosis naming your binding constraint: access, entity, content, or authority

Weeks 3-4: technical and entity fixes

Fix access before anything else, because nothing downstream matters if AI crawlers cannot fetch you: verify every relevant bot reaches your pages, per Technical Accessibility for AI Crawlers. Then fix facts: assemble the brand-facts repository from Running the Program: Workflow, Team, and Governance, reconcile every copy of your facts machines can read, per Entities: Becoming a Thing the Machine Knows, and ship structured data on the pages that matter, per Structured Data and Machine-Readable Facts.

Exit criteria:

• Crawler access verified in logs, not assumed
• One canonical fact set, with site, schema, and major third-party listings matching it
• Stale claims purged from your highest-value pages, per Freshness, Cutoffs, and Temporal Accuracy

Month 2: content and the tracking system

Content work starts from the audit's gaps: restructure the pages your target prompts should retrieve, per Content Architecture: Building for the Chunk, and write the missing ones, per Writing for Retrieval and Citation, through the pipeline stages in Running the Program: Workflow, Team, and Governance. In parallel, graduate measurement from the baseline spreadsheet to a repeatable system on a schedule, per Building a Visibility Tracking System, and connect it to business outcomes, per Attribution and Business Impact.

Exit criteria:

• The first content batch live
• Tracking running on a weekly cadence without manual effort
• A one-page report format your stakeholders have actually seen once

Month 3: authority work and first experiments

Start the slow work now so it compounds while everything else runs: coverage and links, per Digital PR, News, and the Gatekeeper Publishers and Links: The Signal That Refuses to Die, and legitimate community presence, per Reviews, Reddit, and Communities. Then run your first controlled experiment from the citation-gap list in Competitive Analysis: one change, one hypothesis, one measured window, logged.

Exit criteria:

• Authority outreach in motion, with first placements
• At least one experiment completed and written up
• The quarter-two plan drafted from what the experiment showed

The two variants

The sequence is identical for a solo marketer and an enterprise; the cadence and the overhead differ. Solo or small team: do not parallelize. Each phase produces the input for the next, and the failure mode at this scale is six half-finished workstreams. The technical fixes and the baseline are each a focused week, and a working program at day 90 beats an ambitious one at day 30.

Enterprise: the work above is unchanged, and weeks 1-2 gain a second track, stakeholder alignment. Name the owner, per Running the Program: Workflow, Team, and Governance, agree on the scorecard below before the baseline exists, and book the engineering time for weeks 3-4 in week 1, because crawler and schema changes ride release cycles. Governance artifacts, the facts repository and approved-claims library, stop being optional at this scale.

The scorecard

Measure the program against the scorecard, not against answer share. Share can move inside 90 days, most plausibly on searched answers for prompts where your content and rankings improved, but it is noisy, and a quarter is a small sample of a probabilistic system.

What moves in 90 days, and what takes a year

Inside 90 days: access, facts, structure, measurement, and early searched-mode gains where retrieval was the constraint. Inside a year: the compounding returns of authority work, and the first chance for memory-mode movement, because between now and then the providers will train and ship new models, and what the web said about you in the interim is what those models absorb.The memory and retrieval timescales, and the evidence behind them, are in Model Memory and Live Retrieval, drawing on the OppAlerts study documented in The Research Behind This Guide. The concentration data says patience has a payoff: AI answers hold roughly twice the top-five concentration of Google organic, so a slot won here is worth more, and defended more easily, than the ranking it replaces.

Day 91 is the program from Running the Program: Workflow, Team, and Governance on permanent cadence, and Running an AI Search Optimization Campaign is the full agency-grade version of this sequence. OppAlerts runs that process for clients, if you would rather hand it to someone who has run it before.