A 90-Day AEO Roadmap for Teams Starting from Zero

Most teams know they should be doing something about AI visibility but stall on where to start. This roadmap removes that friction. It’s a concrete 90-day plan that takes you from “we have no idea where we stand” to a measured, defensible position — without needing a big budget or a dedicated team on day one.

It maps to the framework in how to build an AEO strategy; think of this as the calendar version.

One caveat up front, because it shapes everything below. Ninety days is a planning convenience, not a natural cycle. Nothing about AI search resolves on a quarterly boundary. What ninety days is long enough for is one complete loop — measure, change something, measure again — and the discipline of closing that loop matters more than the specific length of it.

Weeks 1–2: Establish your baseline

You can’t improve what you haven’t measured, so start by finding out exactly where you stand.

• Define your target query set. Write down 30–50 questions your customers actually ask, across three tiers: category and awareness (“best X for Y”), brand (“is [brand] good for Z”), and use-case or problem queries. This set is the denominator for every metric that follows, so it needs to stay fixed. Swapping queries between measurements silently rewrites your own history.
• Run the baseline audit. Test each query across the engines you care about — ChatGPT, Perplexity, Gemini, Claude, Grok, Meta AI and Google AI Overviews are the seven surfaces multi-engine monitoring tracks. Record presence, position, sentiment, accuracy, citations and competitors. The AI visibility audit guide and the audit checklist walk through the mechanics.
• Calculate your Day 0 numbers: mention rate, average position, accuracy rate, and share of voice.

The part most teams skip

Run each query more than once. Generative answers are sampled, not looked up, so the same prompt can return a different shortlist on two consecutive runs. A single pass gives you a number; it does not give you a baseline.

If you have budget in weeks 1–2 for only one thing, spend it on repetition rather than breadth. Ten queries measured three times tells you more than thirty measured once, because only the first tells you how much of the movement you see later is real. Teams that skip this step spend cycle two explaining variance to their stakeholders as though it were performance.

By the end of week 2 you have a gap map, a baseline, and — importantly — a rough sense of your own measurement noise.

Weeks 3–4: Fix the cheap, high-impact problems

The first wins are usually technical and fast.

• Fix brand-safety issues first. Any inaccurate or harmful claim about your brand outranks all other priorities, because it reaches every user who asks. See why does AI get my brand wrong.
• Clear technical barriers. Check robots.txt for unintended AI-crawler blocks, confirm key pages are crawlable and fast, and add relevant structured data. Note that crawler policy and answer eligibility are different levers. Google’s guidance on AI features and your website states that a page is eligible for AI Overviews and AI Mode if it is indexed and eligible to appear with a snippet, and that no AI-specific markup or new machine-readable file is required. Schema is worth doing for clarity and extraction — not because it buys eligibility you would otherwise lack.
• Tighten your entity. Make your name, category and key facts consistent across your site and third-party profiles. See entity building.

These steps often recover visibility you were losing for purely mechanical reasons, which is exactly why they come before content work. It is demoralising to spend six weeks writing while a stray Disallow line does the damage.

Weeks 5–10: Build citable content for your priority gaps

Now the core work: producing content that earns mentions and citations.

• Prioritize ruthlessly. Start with commercial queries where a competitor appears and you don’t. Those are the gaps with a named beneficiary, which makes them easy to justify internally.
• Write answer-first. Lead each page with the direct answer, use statement headings, and include specific, attributable facts. See how to get cited by AI and writing for AI citation.
• Add FAQ sections to high-intent pages. They mirror how people phrase questions and are straightforward to extract from.
• Publish something original. The 2023 paper GEO: Generative Engine Optimization measured which content edits moved visibility inside generative engines and found that adding quotations, statistics and cited sources helped, while keyword-oriented edits did not. That is one study on one benchmark, not a law of nature — but it is a real measured result rather than an assertion, and it points the same way as practitioner experience.
• Use a citation-first engine as your fast feedback loop. Because Perplexity and Google AI Overviews show their sources, you can see relatively quickly whether new content is being drawn on, and invest more where it is.

Aim for depth over volume. A handful of genuinely authoritative, well-structured pages beats a pile of thin ones, and the thin ones cost you review time you would rather spend elsewhere.

A worked example

Suppose your baseline shows you absent from “best expense management tool for startups” across all seven engines, while two competitors appear consistently. The instinct is to write a landing page targeting that phrase. That is usually the weaker move.

What tends to be doing the work in that answer is the rest of the web’s verdict — comparison articles, review platforms, community threads — not your own page describing yourself favourably. So the higher-leverage sequence is: publish one genuinely useful, specific comparison page, including honest treatment of where you are the wrong choice; make sure your category and pricing facts are consistent everywhere they appear; and pursue coverage on the third-party surfaces already being cited in those answers. Then re-measure that one query set and see whether anything moved.

The reason to write the honest comparison rather than the flattering one is not virtue. It is that engines synthesize across sources, and a page that disagrees with every other source about your weaknesses is a page that reads as promotional and gets treated accordingly.

Weeks 11–12: Measure, report, and plan the next cycle

• Re-run your query set — same queries, same repetition count — and compare against Day 0.
• Attribute carefully. You changed several things at once, so most of what you see is correlation, and you should write it down as such. “We fixed the crawler block and published four comparison pages; mention rate on the affected queries rose” is honest. “The comparison pages drove a 30% lift” is a number you cannot support and will be asked to defend.
• Report it cleanly. Lead with the trend, break results down per engine, and show your coverage as well as your findings — see the reporting template.
• Plan cycle two. AEO is a loop, not a launch. Roll the remaining gaps forward.

What to expect, and what not to

Be realistic about the two timelines. Retrieval-based answers — Perplexity, AI Overviews, any engine calling a live web-search tool — can reflect your work within days to weeks. Knowledge baked into a model’s weights updates only when a new model ships. So expect early, measurable movement on the retrieval side within this first 90 days, with training-driven gains arriving on someone else’s release schedule. See how long does AEO take.

What you should not expect is a clean causal story. Ninety days of a single-team experiment against a shifting external system does not produce one, and any tool or agency that hands you one is overstating its evidence.

The honest version of a good cycle-one result is narrower than most stakeholders want, and more useful: here is where we stand, here is how much of that number is noise, here are the queries where a competitor is winning and we are not present, and here is what we changed. That is a real starting position. A confident-sounding attribution built on one measurement is not.

The point of starting now

AEO is still under-contested in most categories. A team that runs even one disciplined 90-day cycle can establish a measured position — and a repeatable process for improving it — before competitors treat AI answers as seriously as they treat Google. The roadmap above is deliberately modest in scope. That’s the point. Start small, measure honestly, and let the results justify scaling up.