AEO strategy: a 90-day plan for getting named in AI answers

Most AEO strategies are a list of tactics with no order and no way to tell which ones worked. A strategy you can run is shorter: a fixed set of questions, a baseline, work in the order that moves answers, and a verdict on each change. Here's a 90-day version.

The principles

Four ideas carry the whole plan:

1. The questions are the yardstick. Pick them carefully, then leave them alone, or you can't compare one month with the next.
2. One answer is an anecdote. Measure rates over many answers, with a noise band.
3. The order of work matters. Fixing what's wrong beats writing new pages; updating pages assistants already read beats both.
4. Every change gets a verdict. Compare the questions it targeted with ones it didn't touch.

Weeks 1 and 2: questions, rivals, baseline

Choose the questions. 20 to 30 per product line, in buyers' words, about half of them buying questions, with your brand name kept out of those. Sources: sales calls, support tickets, long queries in Search Console, the forums your buyers read. Our guide to choosing the questions to track has the detail.

Choose the rivals. The two to five companies you actually lose deals to, not the biggest names in the industry.

Take a baseline. Ask each question several times on each assistant: ChatGPT, Gemini, Google's AI Overviews and AI Mode, Perplexity. Record named, named first and cited, for you and each rival. Two weeks of answers gives you a starting rate with a usable noise band.

Weeks 1 and 2, in parallel: the plumbing and the facts

While the baseline builds, do the cheap fixes:

• Crawler access. Check robots.txt and your CDN for blocks on the search crawlers (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot, Bingbot). See AI crawlers and robots.txt.
• Wrong facts. Read the answers that mention you and list every wrong price, product, feature or claim. Fix each at its source; our guide to fixing wrong AI information covers how.
• Key facts in text. Pricing, what you sell, who it's for, in plain HTML on the pages that should carry them.

Weeks 3 to 8: work the gaps, one change a week

Rank the questions by the gap between you and the leading rival, weighted toward buying questions. Then, for the top gap each week, make the fix the situation calls for, in this order:

Our guide to getting named in ChatGPT explains why the order works. For each change, write down the date it went live and the questions it targets, and leave the other questions untouched.

One change a week is a sustainable pace for a small team, and it keeps changes far enough apart to tell them apart in the numbers.

Weeks 6 to 13: read the verdicts

Starting three weeks after the first change, read each one: how far did its target questions move, against how far the untouched questions moved, and is the gap wider than the noise band? Read again at five weeks. Our guide to proving AEO results walks through it.

• Worked: do more of that kind of change on similar questions.
• No effect: revise the page once, then move on.
• Hurt: look at what changed; revert if needed.

Week 13: the quarterly review

• Numbers: named, named first and cited, by tier and by assistant, against the baseline, with their bands.
• Verdicts: which kinds of change moved answers for you and which didn't.
• The question set: add questions for anything you've launched; retire the off-target ones; expect a new baseline for anything reworded.
• Next quarter: the next gaps in order, and more third-party work if category questions are where you're losing.

Who does the work

The weekly loop is a few hours: read the numbers, pick the change, draft it, ship it, record it. SEO and content teams can absorb it; for SEOs, much of it is familiar, as our guide to AEO vs SEO explains. Google's own advice is that optimizing for its AI features is still SEO, and that original, first-hand content matters most over time.

It's also a loop an AI agent can run with a person approving each change. Keats does the measurement and the verdicts each week, and hands an agent or a person the next card with its brief.

What to leave out

• Tracking hundreds of questions before you can act on thirty.
• Daily reporting. Answers vary too much; read weekly.
• Tactics with no mechanism, like hidden text for assistants or files promising visibility.
• Changing everything at once. You'll never know what worked.