How to do AEO: the complete playbook, in working order.

Doing AEO means running seven steps in working order: measure who the engines name when buyers ask, diagnose why the winners win, fix the layer machines read, build pages shaped like answers, get indexed where the citations demonstrably come from, earn mentions in the places engines quote, and prove what moved before going again. None of it needs secret access or an enterprise budget — each step below is sized by effort and by what it actually moves.

The playbook at a glance

Working order matters: the most common failure mode is starting at step 4 — writing content with no baseline, no diagnosis, and no index path. The sequence:

1. Measure who the engines name today — about an hour by hand.
2. Diagnose why the current winners win — an afternoon.
3. Fix the technical layer so machines can read you — about a day, most of it once.
4. Build answer-shaped pages — ongoing, one real page at a time.
5. Get indexed in Bing, where the measured citation overlap is — an hour of setup, minutes per page after.
6. Earn third-party mentions — the slow lane, weeks to months.
7. Prove and repeat — an hour a month once the wiring exists.

Short on time? The owner's version, organized by how fast each fix pays off, is how to get your business mentioned by AI — this page is the full workshop manual. If the acronym itself is new, start with what AEO is.

Three ground rules before step one

1. There is no submission form. You cannot register your business with ChatGPT or buy a slot in an organic answer. Every step below changes the inputs the engines read — pages, structured data, indexes, mentions — and the engines decide what to do with them.

2. You win questions, not “AI.” Visibility is per-question: an engine can name you for emergency-type asks and ignore you for best-of asks in the same market. Everything here aims at a specific list of buyer questions — which is why measurement comes first.

3. The board reshuffles monthly. The sources engines cite are not settled furniture:

Step 1 — Measure who the engines name today

Effort: about an hour by hand · What it moves: nothing yet — it makes every other step checkable

Write down ten questions your buyers ask before choosing what you sell, in three shapes: best-of asks, comparison asks, and problem-shaped asks that never name a brand. Then:

1. Ask each question in ChatGPT, Gemini, and Perplexity, in fresh chats.
2. Record three outcomes per answer: were you mentioned, were you cited as a source, were you named first.
3. Write down every competitor named — and the sources shown next to their names.
4. Date the log. Next month's run needs something to compare against.

A faster manual variant — five prompts, one engine — is in does ChatGPT know your business. Or skip the spreadsheet: the free audit runs ten buyer questions across all three engines — 30 live queries — in under three minutes, no account, no email.

Step 2 — Read why the current winners win

Effort: an afternoon · What it moves: your build list — which fights are winnable

The engines show their work more than people expect: Perplexity cites sources on every answer, AI Overviews link theirs, and ChatGPT cites when its live search runs. For every question where a competitor gets named, collect the citations behind the answer and sort them into three piles:

• Pages they own.Contestable by building a better page against the same question — that's steps 3 through 5.
• Third-party sources. Reviews, directories, community threads. Contestable by earning presence — step 6.
• Structural facts. Twenty years of history, a Wikipedia entry. Not contestable this quarter — stop budgeting hours for these.

The full forensic method lives in competitors showing up in AI answers. This step's output is your target list: the questions where the citations are beatable.

Step 3 — Fix the layer machines read

Effort: about a day, most of it once · What it moves: whether engines can read and quote you at all

Four jobs, all quiet, all load-bearing:

• Crawl access — and know which bot does what.The single most common robots.txt mistake in this field is treating “the AI crawlers” as one group. They are not. For appearing in ChatGPT Search the crawler that matters is OAI-SearchBot; block it and you leave that surface. GPTBot is a different agent doing a different job — primarily training-data collection — and blocking GPTBot does not remove you from ChatGPT Search. Anthropic splits the same way: Claude-SearchBot for search appearance, ClaudeBotfor training, Claude-User for a user's own fetch.
• What to actually allow. If search visibility is the goal, allow OAI-SearchBot, Claude-SearchBot, PerplexityBot, and Google-Extended. The training crawlers are a separate, content-licensing decision — plenty of publishers allow search and block training on purpose, and that combination is coherent. Make it deliberately rather than inheriting whatever a template shipped with. Then confirm key pages serve their content as plain HTML rather than only after JavaScript runs.
• Structured data — for parsing, not for favor. Organization or LocalBusiness schema stating name, location, and services; FAQPage schema where visible questions and answers exist. Be precise about what this buys you: Google states that no special markup is requiredto appear in AI Overviews or AI Mode, and there is no such thing as “AEO schema.” What markup does is state your prices, hours, service areas, and Q&A in a form that leaves a machine nothing to interpret — plus it still drives classic rich results. Cheap to add, unambiguous when read. Not a toll you pay for citations.
• The no-cloaking rule. Everything in your markup must also appear on the visible page. A mismatch reads as deception to the engines and can cost you the standing the markup was meant to state.

Step 4 — Build pages shaped like answers

Effort: ongoing — one real page a week beats ten thin ones · What it moves: retrieval-side answers, the pipeline that responds in weeks

An answer-shaped page has one job: fully answer one buyer question. The shape that gets quoted: a question-form heading, a direct answer in the first 40–120 words that stands alone out of context, and specifics an engine can lift — prices, timelines, service areas, who it's for and who it isn't. Comparison questions deserve pages too, and honesty is what makes them quotable: a comparison that admits when the other option wins reads like a source, not an ad.

Step 5 — Submit to Bing, for the reason that survives scrutiny

Effort: an hour of setup, minutes per page after · What it moves: the index with the largest measured overlap with ChatGPT's citations

The step most guides skip — and the one most guides overstate when they don't. Here is the version that holds up. Independent testing found a large overlap between where ChatGPT's citations come from and what ranks in Bing:

Microsoft Copilot is the unambiguous case: it reads Bing's answer layer directly, so Bing standing is straightforwardly Copilot standing.For ChatGPT, treat Bing coverage as a strong leading indicator you can actually act on rather than a gate someone documented. Either way the work is the same and it is cheap: register your site in Bing Webmaster Tools, then submit new and updated pages through IndexNow instead of waiting to be crawled. The honest mechanics of that submission — IndexNow tells Bing a URL changed within minutes; the submission is yours, the verdict is Bing's. A page no index has seen is hard for any retrieval step to reach, however well written. The full engine playbook is Bing, Copilot, and where ChatGPT's citations come from.

Step 6 — Earn mentions where the engines listen

Effort: the slow lane — weeks to months · What it moves: citation breadth, and how the models describe you between releases

Engines quote third parties as readily as they quote you — often more, because a third party saying you're good is evidence, and you saying it is marketing. Three channels, in rough order of yield:

• Reviews. Current and accumulating, on Google and the platforms your industry actually uses. Engines read both the scores and the text.
• Directories.The unglamorous work: identical name, address, and phone everywhere you're listed. Inconsistency reads as ambiguity, and retrieval pipelines route around ambiguity.
• Communities. Reddit threads and niche forums appear in citations constantly. Find the threads engines already cite in your market and participate honestly — the compliant playbook is in why AI cites Reddit so much.

Step 7 — Prove what moved, then go again

Effort: an hour a month once wired · What it moves: next month's build list

Three kinds of proof close the loop:

Answer-side. Re-run the exact question set from step 1 on a fixed cadence and compare run to run: new mentions, lost mentions, whose citations changed. On timing, the honest data:

Traffic-side. Configure analytics to isolate AI-referred visits — the GA4 setup takes an evening. Worth doing before you dismiss small visit counts: in thestacc.com's 2026 data, AI-referred visitors convert at 7% versus 5% and stay 15 minutes versus 8.

Receipts. Log every shipped page with its ship date and target question, and record the date it first appears as a citation. This is the discipline that separates effort from progress — without it, month three is indistinguishable from month one.

The 25-point checklist

The playbook compressed to yes/no, in the same order. Print it, or paste it wherever your team tracks work: every “no” is a task with a step number attached.

The first group is the one you can hand off in minutes: the free audit runs your baseline live — ten buyer questions, three engines, 30 live queries — and reports who was mentioned, cited, and named first, with the three most revealing answers word for word.

Run it yourself, buy the scoreboard, or hire it out

All three are legitimate. The honest triage:

Yourself. Everything above is doable without a vendor. The real cost is cadence: steps 1, 5, and 7 repeat forever, and the loop dies quietly the first busy month nobody runs it. DIY fits when you have more hours than budget and few enough questions to check by hand.

Monitoring tools. Dashboards in the $29–399/month range (published list prices, verified August 2026) automate steps 1 and 7 — the measuring — and are the right pick when you already have a team that ships content and just need the scoreboard. Many in this band monitor without building, so ask what happens to steps 3 through 6 before signing up.

An agency. AEO retainers typically run $2,000–6,000/month (2026 market research). A retainer is the right pick in real cases: a PR-heavy or regulated category where a human strategist earns the fee, an enterprise brand working many markets at once, or a team that wants a person accountable in the room rather than a queue of tasks. What retainers cover — and the questions to ask before signing one — is broken out in what AEO services actually include.

Where our product overlaps this playbook

Full disclosure: AEO Action, which publishes this playbook, built its product as these seven steps run as software — daily scans ask the live engines your buyers' questions; the autonomous department prepares one highest-value artifact a day — schema, FAQ page, landing page, or Reddit pack — and queues it for your approval, and nothing publishes without it; published pages are submitted to Bing; and each approved landing page is checked against its target question for up to 120 days, so a citation arrives as a dated receipt — shipped date A, cited date B. Factor that interest into how you read the triage above. The steps themselves hold whichever tools run them.