Most AI adoption stalls for the same reason: the tooling arrives before the decisions do. This is the order we run it in.
This roadmap covers adoption for marketing and growth functions specifically. It assumes you already have a website, some search presence, and pressure from a board or executive team to "do something about AI". It is sequenced so that each stage produces the input the next one needs.
Why roadmaps fail
Three failure patterns account for most of it.
Tooling before policy. A team buys an AI writing tool, publishes at volume, and discovers later that nobody decided what quality bar applied. Otterly.ai documented two fresh domains that each published roughly 1,000 AI-generated posts and were both algorithmically deindexed by Google with no manual action, one falling from 1,629 to 15 daily impressions. That is a two-site experiment, not a controlled study, and it should still be enough to make anyone pause before scaling unreviewed output.
Visibility before legibility. Teams optimise for AI answers before their own entity is machine-resolvable. Corroboration cannot attach to a company that retrieval systems cannot identify.
No baseline. Work starts, results are claimed, and nobody can prove causation because nothing was measured first. See GEO ROI for the measurement discipline this requires.
Phase 0: readiness, before any spend
Answer these before buying anything.
| Question | Why it gates everything downstream |
|---|---|
| Is our crawler policy deliberate? | Blocking retrieval crawlers makes AI visibility work impossible |
| Is our organisation a resolvable entity? | Without entity clarity, citations attach to the wrong company or nowhere |
| Do we have a baseline? | Without one, no later claim is defensible |
| Who signs off on published content? | Unreviewed volume is a deindexing risk |
| What are we allowed to claim? | Regulated sectors carry constraints that override tactics |
The crawler question is the one most teams get wrong. It is two separate decisions wearing one robots.txt file.
| Decision | Agents | Consequence of blocking |
|---|---|---|
| Training | GPTBot, ClaudeBot, Google-Extended, CCBot |
Content licensing position. No direct search or retrieval penalty |
| Retrieval | OAI-SearchBot, PerplexityBot, Claude-SearchBot |
Removes you from that platform's answers entirely |
| Indexing | Googlebot, bingbot |
Removes you from search, and from AI Overviews and AI Mode eligibility |
Google states that eligibility for AI Overviews and AI Mode requires only that a page is "indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements". There is no separate AI opt-in.
Days 1 to 90: foundations
The first quarter is measurement and legibility, not content volume.
Fix the crawler policy deliberately, documenting the training and retrieval decisions separately.
Establish the baseline. Fix a prompt set that mirrors real buying questions. Sample it repeatedly across the platforms your buyers actually use. Record it before anything changes.
Resolve the entity. Consistent organisation naming, structured data that matches visible text, corroborating references from sources you do not control. Google's own guidance is explicit that structured data must match visible content.
Audit what already exists. Most teams have more usable material than they think, described badly.
Deliverable at day 90: a documented crawler policy, a fixed baseline, a resolvable entity, and a prioritised gap list.
Months 3 to 6: answer engineering
Now content work earns its place, aimed at the questions buyers actually ask rather than keyword volume.
The original GEO research (Aggarwal et al., KDD 2024, arXiv:2311.09735) tested which content changes shifted visibility in generative responses. Reported directionally, the largest gains came from adding quotations from relevant or expert sources, then adding statistics and data points, then rewriting for fluency, then citing sources. Technical terminology and authoritative tone produced smaller effects. Keyword stuffing had a negative effect.
Two caveats matter. The study predates AI Mode and the current model generation, so treat it as directional evidence about how synthesis systems select material, not as a 2026 ranking-factor list. And its authors noted efficacy varies by domain, which is why we do not port tactics between industries unexamined.
Deliverable at month 6: a body of retrievable, quotable, sourced material mapped to real buying questions, plus a first re-measurement against baseline.
Months 6 to 12: corroboration and compounding
Authority is built outside your own domain. Independent references, accurate third-party descriptions, consistent claims across every surface. This is the slowest stage and the most defensible once established, because competitors cannot buy it quickly.
Two things to plan for. Citations decay: Otterly.ai found 19.3% of more than 20 million cited URLs across seven AI engines were dead in a one-month snapshot, highest on ChatGPT at 25.1% and lowest on Google AI Overviews at 12.6%. That is descriptive vendor data, not a controlled study, but link rot at that scale means maintenance is part of the programme, not an afterthought.
And per Profound's sample of 11.84 billion citations across 3.02 million domains and 29 industries between April and July 2026, roughly 57% of AI citations globally pointed to brand-owned domains, ranging from 47% on ChatGPT to 69% on Gemini. Vendor data, large sample, unaudited. If it is directionally right, your own properties matter more than the field assumed.
Governance you should not skip
Decide who reviews AI-assisted output before it publishes. Decide what claims require a source. Decide how often the crawler policy is revisited. Write these down. A roadmap without governance becomes a volume problem within two quarters.
Where this fits
This roadmap maps onto our five-phase methodology: Phase 0 and days 1 to 90 correspond to the AI Visibility Audit and Entity and Authority Foundation, months 3 to 6 to Answer Engineering, and months 6 to 12 to Citation and Consensus Building followed by Monitor, Measure and Compound.
Read the methodology for the operational detail, GEO optimization for what actually moves visibility, GEO ROI for the measurement model, and the glossary for definitions. Our AI Visibility service is where we run this with clients.