Building an AEO Strategy: A 90-Day Operating Plan

An AEO strategy is a sequenced operating plan, not a content list. Days 1 to 30 produce a prompt set and a baseline, days 31 to 60 produce published evidence on domains you do not own, and days 61 to 90 produce the first comparison worth putting in front of a board. Run those phases out of order and you lose the ability to prove anything happened.

Demand Gen Report's 2026 B2B Trends Research Report, published in March 2026, surveyed more than 300 B2B marketers. It found that 72% name a lack of resources as their main barrier to hitting objectives, and that 58% lack the internal resources to produce the volume or variety of content their strategy demands. This plan adds a workstream to teams in that position, so it is built around hours, gates, and named owners.

Key Takeaways

• Sequence carries more weight than tactics: baseline first, publish second, measure third. The order is what makes the month-three number defensible.
• Month one produces nothing public: days 1 to 30 output a prompt set, a recorded baseline, and a source-gap map. Nobody outside the team ever sees them.
• Contrarian: the binding constraint on a DIY plan is not skill or budget. It is naming one person who is still accountable in week nine, after the launch that got postponed.
• Two clocks run at once: retrieval-led visibility can respond to newly published sources sooner, while changes that depend on broader model knowledge may take longer. Do not expect the full multi-engine picture to move at the same speed.
• Gates beat deadlines: end every phase on a check you are allowed to fail. Otherwise, the plan turns quietly back into a content calendar.

What an AEO Strategy Must Cover

An answer engine optimization strategy has to carry five workstreams. It stops being a strategy the moment one of them has no name against it.

Prompt mapping defines what you are measured on, and source-gap analysis names where the work happens. Off-site publishing does the work. The on-site floor stops you from losing on technicalities, and measurement decides what happens next quarter.

The ownership column is what separates this page from a vendor engagement. An agency sells you a schedule of things it will hand over. How the deliverables are paced is the thing you negotiate.

Run it yourself and somebody on your payroll signs off the prompt set. That same person approves a positioning claim before it enters a public thread. In week thirteen, they decide whether the programme continues.

The hours column below is this page's estimate for a two-person team, not measured data. Treat it as a starting model and adjust it against your own calendar before you commit to it.

Days 1-30: Baseline and Prompt Mapping

Nothing in the first month is visible to anyone outside the team. That is the point, and it is also the month easiest to skip.

Building the Prompt Set

Write 20 to 40 questions the way your buyers phrase them, out loud, in a meeting. Four families cover most categories: category questions, comparison questions, problem questions, and brand questions.

Take an industrial and manufacturing SaaS selling maintenance scheduling to mid-size plants. Its buyers do not search for "maintenance scheduling software."

They ask an assistant which system a plant with forty technicians and no CMMS should start with. Then they ask whether it integrates with the ERP they already run. Those are the prompts.

Keep the set small enough to re-run by hand. Twenty-five prompts across five engines is 125 recorded answers a month, which is a morning.

The craft of building a prompt set worth tracking is choosing questions with a commercial consequence. Then you refuse to add more until the quarter turns.

Recording the Baseline

Run every prompt against every engine your buyers use. Paste the raw answer text into a sheet. Summaries are worthless here, because next month you need the actual wording to compare against.

Record four things per answer, every time:

• Presence: were you named at all, and in what position within the answer.
• Description: the exact sentence describing you, including anything wrong.
• Rivals: which competitors appeared, and which appeared first.
• Sources: every domain the engine named or linked.

The gate for this workstream is reproducibility. Hand the sheet and the method to a colleague, and ask them to run five prompts. If their output does not look like yours, you do not have a baseline, and month three will prove nothing.

Mapping the Source Gap

The fourth column is the asset. Sort every domain the engines cited by how often it appeared across the set. Then mark which ones never mention you.

What comes out is usually short and unglamorous. For the maintenance-scheduling example, it might be two review platforms and a trade publication. If relevant subreddit or forum threads appear repeatedly in the source set, add one or two of those communities to the map.

Do not commission a single piece of content until this map names five destinations. The map is the difference between a publishing plan and a wish.

Days 31-60: Off-Site Footprint Building

The second month is where the plan gets expensive in hours. It is also where most in-house attempts stall. Pick fewer places and show up in them consistently.

Choosing the Five Sources

Take the top of the gap map and cut it to five. Weight them by how often the engines cited them. Ignore how much you like the destination.

• Community threads carry the most weight and take the longest to earn. Reddit and specialist forums now behave as discovery infrastructure rather than as a social channel. A plan that ignores them ignores its best surface.
• Review platforms are structured and can appear repeatedly in commercial-category source sets, making absence easy to identify and relatively straightforward to address.
• Comparison listicles move shortlist answers directly, and inclusion is an outreach job rather than a writing job.
• Trade press is slow, high trust, and worth exactly one pitch a month.
• Directories are dull, cost almost nothing, and feed how the engines understand your entity.

Publishing at a Cadence Two People Can Hold

Set the target where you can still hit it in week nine. A pair who commit to twelve pieces a month at kickoff have to hold twelve in the month the launch lands. Leadership remembers the gap between the number promised and the number published.

Community work is the fastest-moving surface in this plan, which is why it starts in week five rather than in the final month. The multi-engine picture can move more slowly. Newly published material can affect retrieval-backed answers before it affects responses that rely more heavily on existing model knowledge.

Everything you publish has to survive being read by a machine. Short, direct answers and plain category language do that. So does structure that lifts cleanly, which is what writing for retrieval rather than for scanning means in practice.

What Not to Publish

Three things waste the second month reliably.

• Owned-domain blog posts that repeat what your product pages already say. The engines can already read your domain, so a restatement adds no new source to the gap map.
• Anonymous community posts written to look organic. Moderators remove them, and the account you burn took weeks to build.
• Anything approved by nobody. A positioning claim published into a public thread without product sign-off is a support ticket waiting to happen.

Days 61-90: Measurement and Iteration

AI visibility optimization becomes a monthly loop on day 61 and stops being a project. The mechanics are boring by design. The comparison only holds if the method does not drift.

Re-Running the Same Prompt Set

Same prompts, same engines, same reporting window, and the same recording method. Change one variable and you have two baselines rather than a trend.

Then compare against the recorded answers rather than against your memory of them. You diff the same four columns you filled in month one: presence, description, rival set, and named sources.

Expect an uneven picture and plan for it. Retrieval-led engines are most likely to move first because they read live sources rather than working from trained recall alone.

Some engines in your set may look untouched at day 90. Read that as an incomplete picture rather than as a failure.

Reading a Result You Can Act On

The useful question is which source moved the number, rather than whether it went up at all. The source is the only part you can repeat next month.

Session variability is real, and the same prompt can return different answers to different users. Treat a single-run change cautiously. A change that holds across repeated runs and coincides with a newly cited source you worked on is a stronger signal worth investigating, not proof that the source caused the movement.

Configure assistant referral tracking now if you skipped it. Assistant referrals arrive as their own source and cannot be reconstructed after the fact.

Add one question to inbound forms asking where the buyer first heard about you. It is crude, and for this channel it beats most analytics.

Deciding What Happens in the Next Ninety Days

Day 90 is a decision, and it needs a stop list as well as a start list. Deprioritise the sources that showed no observable signal during the quarter. Keep prioritising the ones that appeared repeatedly in the engines’ cited source sets.

Write the readout before you write the next plan. One page: what moved, which source moved it, what stayed flat, and what you are stopping. A team that cannot fill that page has run a content calendar for a quarter.

Who Owns What Internally

AEO marketing sits across three functions, and this plan assumes they do not already share a metric. Marketing runs the prompt set. Product marketing owns what may be claimed in public, and sales knows which questions buyers actually ask.

Nothing in the plan fails faster than a workstream with three interested parties and no accountable name.

A Simple RACI

At a two-person marketing team, one person holds three of those columns. Write the names down anyway.

The table earns its keep in week nine. The head of marketing is in a launch, and somebody still has to decide whether a claim goes into a public thread today.

What Not to Delegate

Three decisions stay with the accountable owner. Everything else can move to a freelancer or an intern.

• The identity behind community accounts. Whoever posts is answerable to a moderator and to a customer. That is not a task for a contractor with a shared login.
• Product claims. Anything a model might quote back to a prospect gets read by product marketing before it goes out.
• The stop decision. Killing a source that produced nothing is the hardest call in the plan. It is also the one that keeps the next quarter honest.

Budget and Resourcing

The real cost of this plan is calendar time from two people. The honest way to present it to a CFO is as hours displaced rather than as a new line item.

The hours below are this page's own model rather than measured delivery data. They assume a two-person team and are there to be adjusted against your own calendar.

The phase table above counts hours per workstream and this one counts them per person, so the two totals do not match exactly. Neither says anything about how you split spend across disciplines.

Self-serve monitoring platforms run roughly $29 to $500 a month, on category pricing checked in August 2026. Month one does not need one, because a spreadsheet holds 25 prompts across five engines without complaint.

Buy the tool when the manual re-run starts getting skipped. That is the signal that matters. Read where monitoring stops short of publishing before you assume a subscription closes the gap.

Count the hours properly and loaded in-house cost lands in the $8,000 to $12,000 a month band on the same August 2026 benchmarks. A tool plus a freelance contributor runs $2,000 to $5,000.

Common Strategy Mistakes

Four failure modes are worth designing against. Each has a specific fix that costs an afternoon.

Starting With Content

The commonest sequence error is briefing a writer in week one. It feels like progress and it produces artefacts. It also targets destinations nobody has verified the engines read.

Fix: no brief leaves the building until the source-gap map names five destinations. If that delays the first piece by three weeks, the three weeks were the cheapest part of the quarter.

Tracking Too Many Prompts

A 150-prompt set looks thorough in a kickoff deck and gets re-run once. By month three nobody has touched it, and there is no trend to read.

Fix: cap the first set at 25 and add five per quarter. A prompt you re-run twelve times is worth more than six you ran once.

Judging a Single Engine

Teams check ChatGPT, see nothing, and conclude the programme failed. Engines read different source sets and update on different clocks. One reading is a sample of one.

Fix: report the whole set every month, including the engines that stayed flat. Never let a single-engine spike become the headline either.

Stopping at Week Six

Week six is where the owner gets pulled into a launch and the re-run slips. Nobody kills the programme. It stops having a date attached.

Fix: book the day-90 readout in the calendar during week one. Put the person who signs the budget in the room. A meeting that already exists is harder to cancel than a workstream nobody scheduled.

Free 90-Day Template

This AEO playbook fits into five spreadsheet tabs, and it is deliberately dull. Copy the structure below into a sheet of your own, and take the dates, gates, and owners from the phase table above. That is the whole operating system for the ninety days.

• Tab 1, Prompt set. Columns: prompt, family (category, comparison, problem, brand), buyer role, engines to run, date added, active yes or no.
• Tab 2, Baseline. Columns: prompt, engine, run date, raw answer text, appeared yes or no, position in answer, description of you, rivals named, sources named.
• Tab 3, Source gap. Columns: domain, times cited across the set, mentions you yes or no, access route (community, review, listicle, trade, directory), owner, target date.
• Tab 4, Publishing log. Columns: date, destination, format, author, approved by, live URL, moderation outcome.
• Tab 5, Monthly readout. Columns: month, prompts run, mention count, engines moved, source credited, decision taken.

Tab 5 gets skipped, and Tab 5 is what keeps the programme alive. Six rows a quarter is enough to reconstruct why you did what you did. That is the whole argument in month seven, when somebody asks whether this is working.

When to Hand the 90 Days to Klarivo

Three conditions end the DIY version. Test them in month two rather than discovering them in month five.

• Contributor time. Community contribution needs a real person with standing in a real community, every week, and that person cannot also run demand gen.
• Publishing volume. When the gap map names twelve destinations and your pair can serve four, the plan gets triaged rather than sequenced.
• Sustained measurement. The re-run is the first thing to slip when a quarter gets busy, which makes it the quietest of the three failures.

Klarivo runs this work as a managed programme, with the publishing side and the measurement side both in-house. Klarivo Monitor handles the prompt set, the scheduled re-run, and the trend line across ChatGPT, Claude, Perplexity, Gemini, and Grok. It reports the domains the engines drew on rather than a single score.

The contributor network handles the second month, which is the half that changes what those domains say. Klarivo's Reddit Acceleration page claims real commercial outcomes within one to two months for that managed community work, read there in August 2026. Klarivo is a managed programme for funded B2B brands, and teams too small to use it get pointed at running the plan above themselves instead of into a contract they should not sign.

If your team can hold the hours in the resourcing table, run it yourself and keep the money. That is a real answer, and it is the one this page is built to give.

Not sure which side of that line you sit on? Book a Klarivo discovery call. Fifteen minutes, a slot you choose, and instant confirmation. The measurement side runs in Klarivo Monitor.

Frequently Asked Questions

What does day 91 onward look like once the plan ends?

Quieter, and mostly the same loop on a monthly beat. The prompt set grows by about five questions a quarter, the source list gets pruned rather than extended, and the readout drops from a production to a page. Prioritisation is the part that changes shape. After two comparisons, you know which destinations the engines read for your category, so the publishing plan comes from evidence.

What do you do when an engine describes your product wrongly?

Find the source first. A wrong description comes from something the engine read: an outdated review, a stale directory entry, or a comparison article built from a two-year-old pricing page, and correcting that source is the lever you have. Correcting your own website moves it less than teams arriving from SEO expect. Expect a longer fix than an absence needs, and expect one engine to keep repeating the old line after the others have moved.

Can you run the plan without any paid tooling?

Yes, for the first sixty days. Twenty-five prompts across five engines is a morning of manual work in a spreadsheet, and recording discipline matters more than the software. Manual stops working when you need month-over-month consistency across several engines and a competitor set. Running the same prompts the same way twelve times is what people are bad at, and software is good at.

What is the smallest team that can realistically execute this?

Two people, with one of them accountable rather than merely involved. One person can complete days 1 to 30 alone. The failure comes in the second month, when publishing and measurement compete for the same calendar, and measurement always loses. With one person, run the baseline month, then buy contributor capacity for the publishing rather than pretending the hours exist.

How do you keep leadership patient through the first sixty days?

Show them the month-one artefacts and be honest that no number moved. The prompt set and the source-gap map are real deliverables, and a CFO who has seen the list of five destinations understands why nothing is public yet. Then commit to a date for the first comparison rather than to a result. You control the date, and the result is not yours to promise.