How to Do AEO: A Step by Step Guide for 2026

Most AEO advice stops at the definition. This guide is the opposite: it assumes you already know what Answer Engine Optimization is and want to actually do it, this week, with the site and the team you have now.

The work matters more than it did a year ago. Google reported in July 2025 that AI Overviews had passed 2 billion monthly users, and ChatGPT reached 800 million weekly users by October 2025.

Pew Research Center found that when a Google AI summary appears, people click a traditional result on only 8% of visits, against 15% when no summary appears.

Put plainly, a growing share of your buyers now get an answer without ever seeing a list of links. AEO is the work of making sure your brand is inside that answer.

What follows is a complete sequence: a baseline, prompt research, the page structure that gets extracted, then how to apply it to blog posts, service pages, FAQs and the content you already have.

It ends with the metrics that tell you whether it is working and the mistakes that waste the most time. If you would rather understand the cost and ROI of paying someone else to do this, read AEO services explained.

How do I do AEO step by step for a content strategy in 2026?

Short answer: Work in seven steps: baseline your current AI visibility, research the prompts buyers actually ask, audit which pages can answer them, restructure those pages answer-first, fix entity and technical access, build third-party corroboration, then re-test monthly. Do them in that order, because each step tells you what the next one should prioritise.

The order matters more than the tactics. Most teams start at step four, rewriting content, without knowing which prompts they lose or why, and then cannot explain what changed.

The seven steps

1. Baseline. Pick 20 to 50 real buyer questions. Ask each one in ChatGPT, Perplexity, Gemini, Copilot and Google, and record who gets named and cited. This is your before picture and it takes an afternoon.
2. Prompt research. Expand that list into the full set of questions your buyers ask on the way to a decision, grouped by intent.
3. Content audit. Map each prompt to the page that should answer it. You will find three groups: pages that answer it badly, pages that answer it but bury it, and prompts with no page at all.
4. Restructure answer-first. Fix the buried answers before you write anything new. This is the fastest win available to most sites.
5. Entity and technical. Make sure engines can retrieve your pages and know who you are: crawl access, rendering, schema, consistent brand facts.
6. Corroboration. Build the independent mentions, reviews and discussions that give an engine a reason to trust you.
7. Re-test and iterate. Run the same prompt set monthly against the same engines, and let the losses tell you what to do next.

Steps one to four are realistically a month of focused work for one person on a small site. Steps five to seven are continuous.

Nothing here requires a tool subscription, though tools make step seven much faster.

How do I do AEO from scratch for a new website?

Short answer: On a brand new site, invert the order. Fix technical access and entity clarity first, because nothing else can work if engines cannot retrieve you or do not know who you are. Then publish a small number of genuinely useful answer pages rather than many thin ones, and start building third-party mentions immediately, since that is the slowest part.

A new site has no authority to leverage, which changes the sequencing. You are not restructuring existing equity; you are creating the first reasons for anyone to cite you.

Resist the urge to publish 50 posts in month one. Google's spam policies treat scaled content made mainly to manipulate rankings as abuse, and thin pages rarely earn citations anyway.

Fifteen pages that genuinely answer something beat fifty that skim.

Do not skip the baseline just because you expect it to be empty. Knowing that a competitor is cited for all 30 of your target prompts is exactly the information that tells you which ones are realistically winnable first.

How do I do AEO keyword research for questions people ask AI tools?

Short answer: Research prompts, not keywords. People type fragments into search boxes but ask assistants full questions with context, so your source material is different: sales call transcripts, support tickets, review sites, community threads, and the follow-up questions the assistants themselves suggest. Group the result by intent and test each one to see who is cited today.

A keyword tool will tell you that "aeo agency" has volume. It will not tell you that your buyers actually ask "which AEO agency is worth it for a company our size" and then follow up with "how would I know if it worked".

Those follow-ups are where the decision happens.

Where to find the real prompts

• Your sales calls. The objections and questions that come up before someone buys are almost exactly the prompts they ask an assistant beforehand.
• Your support tickets and inbox. Post-purchase questions reveal the ones that stopped other people buying.
• Review sites and forums. Read how people describe the problem in their own words, not in your category's jargon.
• The assistants themselves. Ask a broad question, then use the follow-up questions the engine suggests. That is a map of how it decomposes the topic.
• Your own search data. Long queries in Search Console are the closest thing you have to conversational intent.

Then group them. A useful set covers definitional questions, comparison questions, cost questions, how-to questions and trust questions, because those map to different pages and different stages.

Google's own documentation on how AI decomposes a search is worth understanding here: one user question often becomes several internal queries, so covering the sub-questions around a topic matters as much as the headline one.

What content structure should I use to do AEO well?

Short answer: Use a question as the heading, a direct 40 to 60 word answer immediately beneath it, then the supporting depth. Keep paragraphs to two or three sentences, use tables for anything comparative, and put facts in their own sentences with a source. That block under the heading is the unit an engine can extract and quote without rewriting you.

This is the single change with the best return, and it costs nothing. Most pages bury their answer in paragraph four after a warm-up about how the industry is changing.

An engine extracting a passage will not find it there.

The pattern, concretely

1. H2 in the buyer's own words. Phrase the heading as the question they asked, not as a clever label.
2. Direct answer first, 40 to 60 words. Complete, self-contained, no throat-clearing. It should make sense lifted out of the page entirely, because that is exactly what will happen to it.
3. Depth underneath. The reasoning, the caveats, the examples. This is what earns the reader's trust once the engine has sent them.
4. A table for anything comparative. Tables are dense, unambiguous and quoted often.
5. One fact per sentence, with a source. A statistic buried inside a clause is harder to attribute and less likely to be used.

The evidence behind the sourcing habit is specific. The Generative Engine Optimization study by Aggarwal and colleagues (KDD 2024) found that adding credible citations, quotations and statistics raised visibility in generated answers by up to 40%, while keyword stuffing did not help at all.

Add structured data on top. Google’s structured data documentation and schema.org cover the types worth having: Organization, Article, FAQPage, Service and Person.

Our free schema markup generator produces valid JSON-LD for the common ones, and our schema guide explains which types matter for AI answers.

What are the best steps to do AEO for blog content?

Short answer: For a blog, consolidate before you create. Merge overlapping thin posts into one comprehensive piece, restructure it so each section opens with a direct answer, add the sub-questions competitors answer and you do not, then keep it current. One authoritative post covering 15 related questions gets cited far more than 15 posts covering one each.

This inverts the usual blog advice. Publishing frequency helps a blog get indexed; depth and structure are what get it quoted.

1. Inventory by topic, not by date. Group every post by the question it answers. Overlap will be obvious immediately.
2. Consolidate the overlaps. Merge into the strongest URL and redirect the rest, so authority concentrates instead of splitting.
3. Add the missing sub-questions. For each surviving post, list the related questions a buyer would ask next and answer them as their own H2 sections.
4. Restructure answer-first. Apply the pattern above to every section.
5. Add a sources list. Real citations to primary sources, not to other blogs.
6. Refresh on a schedule. Engines favour current pages, particularly Perplexity. Dated statistics quietly stop being cited.

Note the redirect step. When you merge posts, the old URLs need 301s so their accumulated authority moves to the surviving page rather than evaporating, and so anyone following an old link still arrives somewhere useful.

Our guide to building topical authority covers the clustering logic in more depth, and content pruning and refreshing covers the consolidation mechanics.

How do I do AEO for service pages so they get picked up by answer engines?

Short answer: Service pages fail at AEO because they are written as sales copy, not as answers. Add a plain description of what the service includes, who it is for, who it is not for, what it costs and how long it takes. Those are the exact things a buyer asks an assistant, and most service pages answer none of them.

Ask an assistant to recommend a provider and watch what it needs: scope, fit, price and proof. A page of value propositions gives it nothing to work with.

The "who it is not for" row is the counterintuitive one. Stating plainly that you are wrong for a segment is unusual enough to be quoted, and it pre-qualifies the enquiries you do get.

Add Service schema and an FAQ block, and link the page to the guides that support it. Our own AEO service page is built this way if you want a working example.

How do I do AEO for FAQ pages without making the content sound repetitive?

Short answer: Repetition happens when FAQs are written as keyword variations rather than as genuinely different questions. Fix it by making each entry answer something materially distinct, keeping answers to 40 to 60 words, and moving anything that needs more than a short paragraph into its own section or page. If two answers could be merged, merge them.

The test is simple. Read two adjacent answers aloud.

If one could be deleted without losing information, it should be.

• One question, one job. "How much does it cost" and "what affects the price" are distinct. "How much does it cost" and "what is the pricing" are the same question twice.
• Answer, do not tease. An FAQ that says "it depends on your needs, contact us" is worthless to a reader and to an engine.
• Vary the shape, not the words. Some answers are a number, some a short list, some a single sentence. Uniform blocks read as generated.
• Promote anything substantial. If an answer needs 300 words, it is a section, not an FAQ entry.
• Use FAQPage schema, honestly. Mark up questions people genuinely ask, not questions invented to hold keywords.

Keep the count sane. Eight to twelve real questions outperform thirty padded ones, and padding is what makes a page sound repetitive in the first place.

What are some practical ideas for doing AEO on existing content?

Short answer: Existing content is where the fastest wins are, because those pages already have authority and indexation. Start with pages that rank on page one but are never cited: they are already trusted and simply structured wrong. Restructuring them answer-first, adding sourced facts and a table, often changes citation behaviour within weeks.

Prioritise ruthlessly. You do not need to touch everything, and the pages worth touching are identifiable in an hour.

A practical half-day exercise: take your ten highest-traffic pages, ask an assistant the question each one is meant to answer, and note whether you are cited. The gap between ranking and being cited is your whole opportunity, and it is usually structural rather than substantive.

How can I do AEO to improve visibility in AI search results?

Short answer: Visibility in AI answers comes from three things happening together: the engine can retrieve your page, the page contains a clean extractable answer, and independent sources corroborate that you are credible. Most brands invest only in the second. Diagnose which of the three is your actual constraint before deciding what to do, because the fix is completely different in each case.

The third row is the most common and least addressed. Google states in its documentation on AI features and your website that these features draw on the same core ranking systems as Search, so classic authority signals still feed AI visibility.

An assistant recommending a provider leans heavily on what other sources say about you.

Engines also differ enough to be worth separate attention. Perplexity always shows sources and favours fresh, well-sourced pages.

AI Overviews lean on core ranking. ChatGPT recommendations are shaped heavily by third-party mentions.

See optimising for AI Overviews and getting recommended by ChatGPT for the per-engine detail.

How do I do AEO for voice search and AI-generated answers at the same time?

Short answer: They are largely the same job now. Voice assistants increasingly return generated answers rather than reading a snippet, so the work that earns an AI citation also earns a voice answer: conversational headings, a direct short answer, plain sentences and clear structure. The one extra consideration is brevity, because a spoken answer cannot be skimmed.

The old voice-search advice about long-tail natural-language phrasing turned out to be the right instinct for the wrong decade. It matters now because that is how people talk to assistants generally, not just out loud.

• Write answers that can be read aloud. If a sentence needs punctuation to parse, shorten it.
• Front-load the answer. A spoken response has no scannable layout, so the first sentence carries everything.
• Keep numbers simple. "About two thousand dollars a month" survives being spoken; "$1,847.50" does not.
• Cover local phrasing if you serve an area. Spoken questions are more likely to include near me and city names.
• Do not maintain two versions. A page that is genuinely clear serves both.

Can you compare different ways to do AEO for local businesses?

Short answer: Local businesses have three viable approaches: profile-led, content-led and review-led. Profile-led fixes Google Business Profile, directories and consistent citations, and is the fastest. Review-led builds volume and recency of genuine reviews. Content-led adds location and service pages that answer city-level questions. Most local businesses should do them in that order.

Local AI answers behave differently from national ones in one crucial respect: they vary by where the person asking is. That means a single national test tells you almost nothing, and you have to test prompts from each market you serve.

Reviews carry unusual weight here, which makes the temptation to manufacture them unusually dangerous. Generating or incentivising reviews without disclosure is now a regulatory matter in the US, not merely a policy one.

Build them genuinely and systematically instead.

Our local SEO checklist covers the profile and citation foundations, and our local AEO service runs all three together for multi-location brands across US and UK markets.

What is the easiest way to do AEO on a small budget?

Short answer: On a small budget, do three things and ignore the rest: restructure your five most important existing pages answer-first, fix your entity basics so engines know who you are, and build genuine reviews. All three are free apart from time, and together they address the most common reasons a small brand is never cited.

Nothing in this list requires a subscription or an agency. The constraint on a small budget is attention, so spend it where the leverage is highest.

1. Week 1: baseline, free. Ask your 20 most important buyer questions across the engines and write down who is cited. An hour or two.
2. Week 2: restructure five pages. Your top five by traffic or revenue. Answer-first, one table, real sources. This is the single biggest win available.
3. Week 3: entity basics. Organization schema, identical brand description everywhere, complete profiles on the platforms that matter in your category. Our schema generator and llms.txt generator handle the fiddly parts free.
4. Week 4: reviews and one genuine community. Ask every happy customer. Join the one forum your buyers actually use and be useful in it.
5. Then repeat the baseline monthly. Same prompts, same engines, same notes.

What to skip on a small budget: content volume, paid visibility tools, and any attempt to compete for prompts dominated by government bodies, marketplaces or major publishers. Pick the specific, winnable questions instead.

Our free SEO and AEO tools cover several of these steps at no cost, and the small-business section of our AEO services guide sets out what is realistic without a retainer.

How do I do AEO and still keep the writing natural and helpful?

Short answer: Use structure to help the reader, not to signal to a model. A question-shaped heading followed by a direct answer is simply good writing; it only becomes robotic when every section is identical in length and rhythm. Vary sentence length, keep a human voice in the depth beneath each answer, and never write a sentence whose only purpose is to contain a phrase.

There is a real tension here, and pretending otherwise is why so much AEO content reads badly. Extractability rewards predictable structure; readers reward variation.

The resolution is to be predictable at the section level and varied within it.

• Predictable openings, varied bodies. Every section can start with a direct answer while the depth beneath differs in form: a story, a table, a list, an argument.
• Keep opinions. A clear point of view is unusual in commodity content, which is precisely why it gets quoted.
• Use real examples. Specifics from actual work are the thing a model cannot synthesise from other pages.
• Cut the transitions. Most connective filler exists to hit a word count. Removing it improves both readability and extractability.
• Read it aloud. The fastest test for robotic rhythm there is.

This is not a compromise with quality; it is aligned with it. Google's guidance on helpful, people-first content emphasises first-hand expertise and genuine usefulness, and AI engines reward the same qualities.

Content written to game extraction tends to fail both.

What metrics should I track to know if my AEO is working?

Short answer: Track share of answer as your primary metric: how often you are cited across a fixed prompt set, per engine, re-tested on a fixed schedule. Support it with AI referral traffic, branded search volume, and self-reported attribution from your lead forms. Do not judge AEO on organic sessions alone, because most AI influence produces no click.

This is where most self-directed AEO programs go wrong. They look at Analytics, see flat sessions, and conclude nothing is happening while their brand is increasingly being named in answers.

Define your terms before you start counting. A mention, a linked citation and an explicit recommendation are three different outcomes, and conflating them makes your numbers meaningless over time.

Expect variance. The same prompt can return different answers on consecutive runs, so test each prompt several times and treat the trend as the signal rather than any single result.

Ahrefs found that AI Overviews reduce clicks to top-ranking pages, which is the pattern that makes click-based measurement misleading here.

What are the top mistakes to avoid when learning how to do AEO?

Short answer: The most costly mistakes are starting without a baseline, writing new content before fixing existing pages, optimising only on-site while ignoring third-party corroboration, chasing volume over depth, treating AEO as separate from SEO, and judging the whole program on organic sessions. Each one wastes months before the problem becomes visible.

One more worth naming separately: publishing answers you cannot support. An engine that quotes a claim you invented will cite you until someone checks, and in regulated categories that is a compliance problem rather than an embarrassment.

How Web of Picasso runs this

Everything above is the process we use, in the order we use it. We publish it in full because the bottleneck for most teams is not knowing the steps, it is having the time and the measurement infrastructure to run them every month.

Every engagement opens with the step most people skip: a share-of-answer baseline across ChatGPT, Perplexity, Gemini, Copilot and Google AI Overviews, so you know exactly which prompts you lose and to whom before anything is written.

We then run content, entity, technical and off-site work as one program alongside SEO, and re-test monthly.

If you would rather do it yourself, use this guide and our free tools, including the schema markup generator, the llms.txt generator and the content analysis tool.

If you would rather not, our AEO services page sets out what we actually deliver each month, and our agency selection guide gives you the scorecard to judge us and everyone else with.

Frequently asked questions

How do I do AEO step by step?

Baseline your current AI visibility across a fixed prompt set, research the questions buyers really ask, audit which pages can answer them, restructure those pages answer-first, fix entity and technical access, build third-party corroboration, then re-test monthly. Do them in that order, because each step tells you what to prioritise next.

How long does AEO take to work?

Expect first citations on niche and long-tail prompts within 60 to 90 days, and broader share-of-answer gains between four and nine months. Restructuring pages that already rank is the fastest lever and can change citation behaviour within weeks; building third-party corroboration is the slowest part.

Can I do AEO myself without an agency?

Yes. The highest-leverage steps cost only time: baselining a prompt set, restructuring your most important pages answer-first, fixing entity basics with schema, and building genuine reviews. An agency mainly adds speed, off-site PR capability and measurement infrastructure rather than secret techniques.

Is AEO different from SEO?

They overlap heavily and should be run together. SEO optimises pages to rank in a list and earn clicks; AEO optimises your brand to be named inside a generated answer. Google states that its AI features draw on the same core ranking systems, so search strength still feeds AI visibility.

What content structure works best for AEO?

A question as the heading, a direct 40 to 60 word answer immediately beneath it, then supporting depth. Keep paragraphs to two or three sentences, use tables for comparisons, and put each fact in its own sentence with a source so it can be attributed cleanly.

How do I measure whether AEO is working?

Track share of answer across a fixed prompt set per engine, re-tested on a schedule, and log mentions, citations and recommendations separately. Support it with AI referral traffic, branded search volume and self-reported attribution, because most AI influence produces no click at all.

Does AEO work for local businesses?

Yes, and the sequence differs. Fix Google Business Profile, directories and citation consistency first, build genuine reviews second, then add city and service pages that answer local questions. Test prompts from each market you serve, because AI answers vary by the location of the person asking.

What is the biggest AEO mistake?

Starting without a baseline. Without a record of which prompts you lose and to whom, you cannot tell progress from normal variance, you cannot diagnose why a competitor is cited instead of you, and you can never prove what the investment achieved.

Sources and further reading

• Pew Research Center (2025): Google users are less likely to click on links when an AI summary appears
• Aggarwal et al. (KDD 2024): GEO: Generative Engine Optimization
• Ahrefs (2025): AI Overviews reduce clicks to top-ranking pages
• Google Search Central: AI features and your website
• Google Search Central: creating helpful, reliable, people-first content
• Google Search Central: spam policies for Google web search
• Google Search Central: introduction to structured data markup
• Alphabet Q2 2025 CEO remarks: AI Overviews reach 2 billion+ monthly users
• TechCrunch (Oct 2025): ChatGPT reaches 800M weekly active users
• schema.org: structured data vocabulary
• llmstxt.org: the llms.txt proposal

Want the baseline before you start?

Step one of this guide is the one most people skip. We will run it for you free: your real buyer prompts tested across ChatGPT, Perplexity, Gemini, Copilot and Google AI Overviews, with who gets cited today and why. See our Answer Engine Optimization (AEO) services or book a free AI visibility audit.