The 90-Day AI Visibility Plan: A Week-by-Week Roadmap

ClickRadius Institute · April 15, 2026

Strategy needs a calendar: what to inspect, what to prepare, what requires approval, and what evidence marks completion. This 90-day sequence covers a baseline, publisher-controlled evidence, governed public records, and later measurement. AI Overview coverage varies by source, cohort, date, locale, device, and definition. The calendar is a work plan, not evidence that a business will enter a cited group by day 90.

Customer planning evidence boundary. No eligible closed ClickRadius customer-month cohort establishes a typical setup or maintenance time, labor saving, payback, return on investment, or citation outcome for a small business. Any sequence or cadence on this page is a planning framework, not a measured completion or result window. Measure actual work, accepted site scope, costs, and attributable outcomes before deciding whether a tool is economical.

Ground rules before week one

• One owner. The plan survives contact with reality only if a single person owns the calendar, even when specialists execute pieces of it.
• A question set is your north star. Everything is measured against 25–50 real buyer questions in your category. You will build this in week 1 and re-test it every month.
• Evidence over adjectives. The Princeton-led GEO study (KDD 2024) tested quotations, statistics, source citations, and other interventions in a bounded benchmark. It began with the top five Google sources, randomly selected one source for optimization, and measured that source's relative visibility/impression in generated responses; it did not measure retrieval or citation-inclusion probability, and reported effects varied by domain. Use it to form testable editorial hypotheses, not universal rules.

A calendar can bound the work to inspect and the evidence to retain. It cannot set the timetable of an external engine or prove that later maintenance will be routine.— ClickRadius Institute

Phase 1 — Baseline and access (weeks 1–2)

Week 1: measure where you actually stand

• Draft your question set: 25–50 conversational questions a buyer would ask, from “best [service] near [city]” to “is [your brand] worth it.” Mine sales calls, support tickets, and search-query reports.
• Run every question across ChatGPT, Gemini, Perplexity, Claude, and Grok. Log three columns per engine: mentioned? cited? who won instead?
• Compute the result for this disclosed sample, keeping no appearance, mention, linked citation, and recommendation separate. Report the numerator, denominator, engines, prompts, conditions, and date.

Week 2: audit the machine-facing site

• Check robots.txt and CDN/firewall rules for each relevant crawler or control token, using current provider documentation to distinguish training, search, and user-request purposes. A declared block proves only the policy for that identified agent or token.
• Inventory structured data: Organization, Article, FAQPage, and LocalBusiness/Product schema coverage.
• Grade your top 20 pages on answer-first structure: does the first screen of each page state a complete answer an engine could lift?
• Verify entity consistency — identical business name, description, and core facts across your site, Google Business Profile, and major directories.

Phase-1 exit criteria: question-set record complete; relevant crawler policies tested; public responses retained; and a prioritized defect list. ClickRadius's free AI Readiness Score estimates five evidence families at the submitted URL. It is separate from the monthly AI Presence sample and is not a completion-time or outcome measure.

Phase 2 — On-site foundation (weeks 3–5)

• Week 3: Fix the structural defects — schema deployment, crawler access, broken or contradictory entity facts. Boring, decisive work.
• Week 4: Review a bounded set of priority pages. Make opening answers accurate out of context, use headings that help readers navigate, and add only the statistics or questions the page can support.
• Week 5: Continue through the accepted inventory. Add quotations only where their speaker, wording, source, and context can be verified.

Retrofitting before creating is an operational sequencing choice: existing pages may contain content and links worth preserving. The GEO study tested selected-source visibility, not whether retrofits are a highest-leverage move or whether live-retrieval engines reflect them within weeks.

Phase-2 exit criteria: the accepted page inventory has a documented state; approved revisions are publicly verified; schema is validated where applicable; and the readiness score is re-run without assuming it must improve.

Phase 3 — Citable-content sprints (weeks 6–9)

Use the week-1 sample and first-hand customer questions to define a bounded content queue. Rank candidates by business relevance, factual support, and observed gaps. The benchmark below can inform tests, but it does not supply a required article count, sprint length, or universal acceptance formula:

In a benchmark beginning with the top five Google sources, researchers randomly selected one source for optimization. Quotations, statistics, and source citations produced reported relative visibility/impression gains of up to roughly 40% in some settings. The experiment did not measure retrieval or citation-inclusion probability, and effects varied by domain.— Princeton “GEO” study (KDD 2024), finding paraphrased

1. One supportable scope per piece. Open with an accurate answer and retain the qualifications needed to interpret it.
2. Evidence as the acceptance test. Verify each statistic, quotation, source, and FAQ answer; do not add arbitrary counts to satisfy a template.
3. Link where useful to readers. Connect genuinely related pages without claiming that a particular link structure compels engine treatment. (Our guide to topic clusters and pillar pages for GEO covers the architecture.)

Phase-3 exit criteria: each accepted piece has factual and editorial approval, a delivery receipt, and independent public-origin readback. External-engine use remains a separate later observation.

Phase 4 — Govern off-site records (weeks 8–12 in this work plan)

This is an internal delivery schedule, not an AI-outcome clock. Run the off-site evidence track separately from first-party page work and preserve each provider's acceptance and public-effect states:

• Week 8: Inventory applicable directories, official records, Google Business Profile, review sites, and association listings by owner, exact entity, location, fact scope, and correction path. Do not submit everywhere or force lexical identity.
• Week 9: Establish a policy-compliant request and response process for genuine reviews. Do not set sentiment, volume, or recency as a promised engine-recommendation lever.
• Weeks 10–12: Pursue only truthful earned coverage and publish original data only with method, population, period, limitations, and approval. A pitch is not a published mention.

Phase-4 exit criteria: every scoped source is classified as not applicable, untouched, submitted with receipt, publicly verified, conflicting, failed, or unknown; the review process is documented; and only actually published third-party records are counted.

Phase 5 — Instrument and re-measure (weeks 12–13)

• Re-run the full question set across all five engines. Compare mentions, citations, and citation share against week 1, engine by engine.
• Segment attributable AI-surface referrals where evidence exists, while reporting branded search, inquiries, and conversions separately. Do not assume untracked buyers were AI-influenced.
• Write the day-90 report: AI Readiness observations, separate SEO Health, monthly AI Presence comparison, verified public revisions, provider receipts, public readbacks, failures, and unknowns.

Staffing the plan: who does what

The 90-day plan assumes no dedicated GEO hire, because almost nobody has one yet. The work decomposes cleanly across roles most organizations already have:

• The owner or marketing lead: owns the accepted scope, prioritization, approvals, first-party facts, and interpretation of business evidence.
• A writer, in-house or contracted: prepares factual drafts and preserves attribution, source context, and unresolved review questions.
• A developer or webmaster: handles the technical work actually required by the site's templates, crawler controls, rendering path, and approved schema changes.
• Customer-facing staff: records how customers say they found the business and runs a policy-compliant review-request process without scripting sentiment.

ClickRadius can estimate readiness inputs, prepare eligible content and site revisions, and record a monthly AI Presence sample. An exact site effect still requires current authorization and a connection proven capable of that effect; human review, exceptions, off-site work, and public verification remain part of the workload. Measure actual hours rather than assuming a reduction.

The weekly scorecard

Momentum dies in the dark, so track five numbers on one sheet, updated every Friday:

1. Readiness score (or a documented manual assessment) — record the value and evidence without prescribing a trajectory.
2. Pages passing the density test — cumulative; the plan's cleanest effort metric.
3. Question-set coverage — the share of your 25–50 questions with a live, definitive answer page.
4. Answer observations — no appearance, mention, linked citation, or recommendation, by engine and sample conditions.
5. Off-site assets banked — profiles completed, reviews gained, mentions earned. Cumulative and boring, which is the point.

Some inputs are publisher-controlled, some records are provider-controlled, and every answer is engine-controlled. Report those authority boundaries with the observations rather than treating one number as proof of another.

What honest expectations look like at day 90

By day 90, report only what the evidence shows: access tests, completed and publicly verified work, readiness observations, provider receipts, and any mentions or citations actually captured in the repeated eligible panel. No reviewed cohort supports an expected first-citation picture, a compounding quarter, or a three-cause diagnosis for a flat result. Preserve unavailable and unexplained states and investigate them without guessing.

Frequently asked questions

Is 90 days really enough to see AI-search results?

Ninety days defines this operating calendar; it is not a measured completion or result window. Access and completed publisher-controlled work can be verified directly, while mentions and citations may move, remain unchanged, or be unavailable. No external-engine outcome is promised for day 90 or a later quarter.

How many hours per week does the 90-day plan require?

No eligible closed ClickRadius customer-month cohort establishes a typical weekly workload or labor saving. Estimate from the accepted URL, content, provider, approval, and verification scope; record human review and exception work as well as automated steps; then compare actual hours with the estimate.

What should I do on day 91?

Continue the governed maintenance loop: repeat the same eligible five-engine sample monthly, review content on a cadence tied to factual change, and preserve public-record receipts and rereads. Adjust work from observed evidence. Consistency improves the audit trail; it does not make an engine trust the business by default or prove a compounding advantage.

Want a site-input baseline first? Get your free AI Readiness Score — a 0–100 estimate across five evidence families — or see ClickRadius plans for separate monthly AI Presence samples across ChatGPT, Gemini, Perplexity, Claude and Grok. ClickRadius may prepare content and on-site revisions, but it executes an exact revision only with current authorization and a connection proven capable of that effect; otherwise the work remains for review or customer implementation. Neither the score nor this 90-day sequence promises an external-engine outcome.