- This roadmap is the concrete week-by-week version of the five stages for winning in the AI search era. Nothing here is a new framework, it's a realistic timeline laid over one already published and sourced.
- Stage 1 (access) takes days to confirm. It's a mechanical check, not ongoing work. Stage 2 (entity) is mechanically fast but propagation timing isn't rigorously measured anywhere; budget weeks, not days, for it to settle. Stage 3 (content extractability) is a real multi-week production effort on cornerstone pages. Stage 4 (earned media) is explicitly a quarters-long, compounding process: 90 days starts it, doesn't finish it.
- The one genuine fast exception: Kevin Indig's G2 research found a median four-day lag between a real review burst and a measurable citation lift, a real early signal achievable inside this window, not evidence the full off-site stage is done.
- Monitoring (Stage 5) gets set up early, in week one or two, alongside Stage 1, specifically so there's a real baseline in place before the other stages start moving anything, rather than being bolted on at the end.
- No independent, disclosed-methodology study shows a durable AI-citation lift achievable on a fixed 90-day timeline net of brand size and existing review volume. Anyone selling that outcome on a guaranteed clock is selling past what the evidence in this framework actually supports.
Ninety days is a reasonable planning horizon for the foundation. It is not a reasonable horizon for finishing the whole five-stage sequence, and a roadmap that implies otherwise is exactly the kind of guaranteed-timeline pitch this site has already flagged as unsupported by any disclosed study. What follows is a realistic week-by-week plan: what's actually achievable in 90 days, what genuinely can't be rushed, and where the honest boundary between the two sits.
What does the roadmap look like, at a glance?
| Weeks | Stage | What's actually happening |
|---|---|---|
| 1 | Stage 1 — Access & indexation | Confirm crawlers can reach the site; fix anything broken |
| 1-2 | Stage 5 setup — Monitoring baseline | Stand up the tracking that runs for the rest of the 90 days and beyond |
| 2-4 | Stage 2 — Entity consistency | Canonical naming, schema, Wikidata (if eligible), profile alignment |
| 4-8 | Stage 3 — Content extractability | Rewrite cornerstone pages for direct answers, original data, self-contained sections |
| 6-13 | Stage 4 — Off-site & earned (started, not finished) | Review generation, PR outreach, profile hygiene on review platforms |
| Ongoing from week 2 | Stage 5 — Monitoring | Fixed prompt set, run weekly, tracked as a trend from the baseline set in weeks 1-2 |
Week 1: is the gate actually open?
Everything else in this roadmap is wasted spend if this fails. Run the full three-layer access check (robots.txt, server logs, and the CDN/WAF layer that can silently override both) and specifically verify the identity of what's actually hitting the site using the crawler-audit methodology, not just the user-agent string. If the site sits behind Cloudflare or a similar CDN, check for the exact silent-override failure this site itself hit before assuming a permissive robots.txt means anything. This step takes under 15 minutes to run and should return a clean result before spending a single hour on anything below it.
Weeks 1-2: should you stand up monitoring before there's anything dramatic to see?
Counterintuitively, monitoring starts almost immediately, not because there's meaningful data yet, but because AI visibility KPIs only work as a trend, and a trend needs a baseline. Build or configure the fixed prompt set now, run it once to establish where things stand before any other stage has had a chance to move anything, and schedule it to run weekly from here forward. Setting this up in week one, not week twelve, is what turns "did this work" into an answerable question by day 90 instead of a guess.
Weeks 2-4: does the entity resolve cleanly?
Run the full entity audit. It's mechanically under an hour of work, but the propagation timing after fixing name inconsistencies, filling out schema, and updating third-party profiles isn't rigorously measured anywhere in the industry, so budgeting into week 4 for things to actually settle is realistic, not padding. If genuinely eligible, set up Wikidata in this window. Check the real qualification bar first, since it's lower and differently-shaped than most guidance assumes, but still a real bar. If G2 or Capterra profiles are part of the entity footprint, bring them to full completeness in the same window: category accuracy and field completeness are cheap fixes that directly reinforce the same entity-resolution signal.
Weeks 4-8: is the highest-value content actually extractable?
This is the one on-site lever with genuine controlled-study evidence behind it, and it's real production work, not a checklist. Budget a full month. Identify the pages that matter most (the ones a prospect would actually land on from an AI-generated answer) and rewrite them using the actual documented patterns: a direct answer near the top of a section, original data instead of a fifth-hand paraphrase, self-contained sections that don't require stitching context from elsewhere on the page. If a deeper understanding of the underlying research helps make the case internally for the time this takes, the GEO paper itself is worth the read, including its own finding that keyword density and length for their own sake actively underperform, which is worth knowing before a well-meaning stakeholder asks for "more content" instead of better-structured content.
Weeks 6-13: which stage doesn't finish in 90 days?
This is the stage to be honest about. Earned media accounts for 82-89% of AI citations across every disclosed-methodology study that's measured it, and it's also explicitly the slowest-compounding stage in the whole sequence, not something a 90-day window completes. What a 90-day window can do is start it properly: begin PR outreach and pitching, and start a genuine review-generation push on the platforms that matter for the category. The one real fast signal available here is Kevin Indig's G2 research, which found a median four-day lag between a genuine review burst and a measurable citation lift, a real, achievable early win inside this window, and a useful signal that the sequence is working, but it's a leading indicator on one platform, not evidence the broader earned-media stage is complete.
What does day 90 actually look like, honestly?
By day 90, on a well-run version of this roadmap: the access gate is confirmed open and monitored, the entity resolves cleanly across every profile that matters, the highest-value pages are genuinely extractable, and the earned-media engine is running with early signal (likely from reviews, per Indig's four-day finding) but nowhere near mature. That's a real, meaningfully different position than day zero, and it's also not "AI visibility achieved." The stages that compound slowest are the ones that matter most for durable citation share, and they're specifically the ones that can't be compressed into a fixed window regardless of budget. A roadmap (or a vendor) that promises otherwise on a guaranteed 90-day clock is making a claim no disclosed study in this space actually supports.
Frequently asked questions
Can you actually achieve strong AI visibility in 90 days?
You can complete the foundation (access, entity resolution, and a real content-extractability pass) well within 90 days, since that work is mechanical and largely under your control. What you can't do in 90 days is finish Stage 4, off-site and earned presence, which this site's own research already documents as a quarters-long, compounding process. A 90-day roadmap gets the gate open and the compounding engine started. It doesn't finish the compounding.
What should happen first in a 90-day plan?
Stage 1 (confirming AI crawlers can actually reach the site) in week one, before anything else. It's a binary gate, not a lever: a brand with perfect entity hygiene and a content overhaul in progress still produces zero citations if a crawler is silently blocked, which is exactly the failure this site's own domain hit once.
Is 90 days long enough to see review-driven citation lift?
For an initial signal, yes. Kevin Indig's G2 research found a median of four days from a genuine review burst to a measurable citation lift, which is a real, fast feedback loop early reviews can produce inside a 90-day window. That's a leading indicator the sequence is working, not evidence the full off-site stage is complete by day 90.
What's the biggest mistake in a rushed AI visibility timeline?
Starting Stage 4 or Stage 5 before Stages 1-3 are actually confirmed solid: paying for earned-media outreach while the site is still crawler-blocked, or standing up a monitoring dashboard before there's a resolved entity or extractable content worth tracking. The stages are a dependency chain; running them out of order is where budget gets wasted, regardless of how good any individual tactic is.
Sources referenced
- Kevin Indig / G2 review-burst citation-lift study (Jun 2026, 2,514 products)
- Muck Rack Generative Pulse, "What Is AI Reading?" (Jul 2025 / Dec 2025 / May 2026 editions, 25M+ links)
This is the concrete timeline built directly on the five stages for winning in the AI search era. Every stage above links back to a full standalone guide for the actual work, not just the schedule.
