The rest of this book explains the work. This chapter walks through how the work gets packaged and run as an engagement: intake, research, a plan, execution, and reporting, in that order. It is the process OppAlerts runs for clients, documented in the open, and it applies whether you are an agency running it for someone else or an in-house team running it for yourself.OppAlerts offers this engagement as a service; the research behind the method is published free as the AI Search Visibility research.
The sequence exists to prevent one specific failure: execution that starts before anyone knows which prompts matter, which source, model memory or live retrieval, produced the current answers, or what number will prove the work paid.
Onboarding and discovery
Discovery answers two sets of questions: what the business actually sells, and what an AI answer about that business is currently made of. The business side comes from the client: products and services ranked by margin, the customer segments that matter, the competitors they lose deals to, the claims they can support, and the constraints, legal, brand, or regulatory, on what can be published. Get the client's beliefs about their AI visibility on record too, because the baseline usually contradicts some of them, and showing that contradiction with data is the first deliverable that builds trust.
The machine side is measured, not asked. Build the prompt inventory with the process in Prompt and Topic Research, then run the baseline: repeated API runs per prompt, search on and off, across the models that matter, as specified in Measuring AI Search Visibility. The memory-versus-retrieval split in that baseline decides most of what follows, because work aimed at model memory and work aimed at live retrieval run on different clocks; Model Memory and Live Retrieval covers why.
Onboarding also sets the working agreement: who approves content, who has site and analytics access, and how fast the client side can actually ship. A plan the client cannot execute at their real shipping speed is decoration.
Research and planning: the core deliverable
In the OppAlerts model, research and planning is the engagement's center of gravity. The standard to hold the plan to: a competent team can execute it without the person who wrote it.
Three research passes feed it. The diagnostic pass runs the gates in Auditing and Diagnosing Visibility Problems to establish why current visibility is what it is. The competitive pass, from Competitive Analysis, identifies who holds each slot the client wants, whether that hold rests on memory or retrieval, and where the unstable prompts, ignored personas, and open sources are. The strategy pass, from Building Your AI Search Strategy, scores the category itself: concentration, persona sensitivity, and the memory-retrieval mix, which together set the budget split and the honest timeline.
The plan that comes out has a specific anatomy: a prioritized asset list (pages to create or restructure, facts to make machine-readable, gaps against the sources the models credit), a promotion list (the publishers and communities the models read in this category), a technical punch list from Technical Accessibility for AI Crawlers, a paid recommendation where the economics support one, and per-item owners and dates. Every item traces to a prompt it is expected to move and the mechanism it moves it through. An item that cannot name its prompt and mechanism gets cut.
Execution: assets, promotion, paid
Execution runs the plan in dependency order. Technical access first, because nothing downstream works if crawlers cannot fetch the site. Content second: creation and restructuring per Content Architecture: Building for the Chunk and Writing for Retrieval and Citation. Promotion third and continuously: the coverage and community work in Digital PR, News, and the Gatekeeper Publishers, which is the slowest work and therefore starts as early as the plan allows. Paid runs in parallel where the plan calls for it, using the decision framework in Paid AI Search Strategy and Tactics.
Whether the planning team also executes is a business-model choice. OppAlerts hands off: research and planning as the deliverable, execution by the client's team or their existing vendors, with the plan written to survive that handoff. If you run execution in-house instead, the workflow, ownership, and QA structure in Running the Program: Workflow, Team, and Governance is the operating manual. For sequencing at the week level, the campaign's first quarter is already written out as The First 90 Days; the process in this chapter is that plan with a client relationship wrapped around it.
Reporting and the operating cadence
Reporting keeps the work funded, and it fails when it promises the wrong resolution. Answer share moves on weeks and months, so the cadence is weekly sampling, monthly reporting, quarterly strategy review. A daily AI-visibility report is temperature noise with a logo on it.
The monthly report carries four things: visibility metrics against baseline (mention rate, recommendation rate, citation rate, and share of voice, by model and prompt class, search on and off), the work shipped, the business-impact picture from Attribution and Business Impact, and next month's priorities, with anything the plan got wrong stated plainly. Model updates go in a version log inside the report, because a provider shipping a new model can move every number on the page with no action from anyone, and a report that cannot distinguish a model update from campaign impact will eventually claim credit for weather.
The quarterly review re-litigates the plan itself: rerun the category scores, rescore the competitive map, and move budget between memory work, retrieval work, and paid based on what the two clocks actually delivered.