AEO Strategy: The 4 Decisions That Decide If AI Names You

An AEO strategy is four decisions, made in order: which buyer questions you compete on, whether you work on model memory or live retrieval, whether your proof lives on your pages or on other people's, and the one number you judge it by. Schema, FAQs and answer-first copy are tactics that come after.

AEO (answer engine optimization, also called generative engine optimization) means getting named in answers AI systems generate. For the basics, read what AEO is.

What separates an AEO strategy from a tactic list?

A strategy makes binding choices about where you compete and what you give up; a tactic list doesn't. Ask what a plan tells you to stop doing: if the answer is nothing, it's a to-do list.

Decision 1: Which buyer questions will you compete on?

Pick 10 to 20 questions a buyer asks when close to choosing, such as "best [category] for [use case]", "[competitor] alternatives" and "is [brand] worth it". Source them from sales calls, support tickets and Search Console, and rank them by how much a mention would change a purchase, not by search volume.

An answer has few slots. In AEOeye's audit records (127 audits, June to September 2026; self-selected brands, not a market estimate), the average answer named 2.9 competitors, and when the audited brand was named and ranked it came first in 71.4% of 147 cases (aggregates).

Definitional questions ("what is [category]") are the weakest picks: they rarely precede a purchase, and engines often answer them from memory, which no page you publish can change this quarter. Challengers should narrow by segment: unbranded queries tend to default to market leaders (Chen et al., 2025), so "best CRM for freelancers" is winnable where "best CRM" is not.

We learned this on ourselves. Search Console showed about half of our 496 posts were unrelated to AI visibility and had no clicks in 90 days. We retired 234 and merged 27 duplicates on 2026-09-24, then merged 165 duplicate-intent pages into 42 on 2026-10-03. This guide replaces two pages that answered the same question. We claim no measured lift yet, only a position: one strong page per question beats five thin ones.

Decision 2: Are you working on model memory or live retrieval?

Work on retrieval first. Retrieval, or retrieval-augmented generation, means the engine searches the web while it answers and cites what it finds; memory means it answers from training data alone. You can change retrieval this quarter; memory changes only when a model is retrained.

The engine decides when to search: Anthropic's documentation says Claude searches for current or changing information, such as products that may have changed, and answers stable knowledge directly (Claude docs). As of October 2026, OpenAI documents separate bots for ChatGPT search (OAI-SearchBot, whose opted-out sites are not shown in answers) and training (GPTBot), and Anthropic and Perplexity likewise keep search crawling apart from training.

Access is rarely the problem. In AEOeye's crawler access study (October 2026, 384 sites, a convenience sample skewed to established brands; method and data), 4.2% blocked OAI-SearchBot and 9.1% GPTBot on the homepage, and 1 of 193 SaaS sites blocked any AI retrieval bot, though 27 of 38 publishers did. Firewalls are the likelier blind spot: 24 of 88 e-commerce homepages (27.3%) challenged or blocked our self-identified research crawler.

Decision 3: Do you build owned pages or earn proof elsewhere?

Earn proof elsewhere first, and use your own pages for facts nobody else will verify. AI engines cite independent reviews, comparisons and publishers far more than brand sites, and they assemble comparison answers from many searches (Google calls this query fan-out), so most pages behind an answer are not yours.

In a September 2025 arXiv preprint, Chen et al. compared a web-enabled GPT with Google on 1,000 consumer ranking prompts. In the software-products test (Canada), 74.2% of the AI engine's citations went to earned media and 25.8% to brand sites; Google's mix was 31.8% earned, 53.8% brand. The authors call the percentages illustrative; earned media led in every vertical and region tested.

So start from the answers, not your site. Run your question list, note the third-party pages the engines cite, and target the five to ten that recur. Publishers restrict AI crawlers most (see Decision 2), so work from citations you actually see.

For a marketing team, that moves budget from content production toward PR, reviews and comparison coverage; Chen et al. likewise advise investing in PR and media outreach. How to get your business recommended by AI covers outreach.

Your own pages still have three jobs:

• State facts no reviewer will check: pricing, limits, integrations.
• Keep your name and category identical everywhere, Organization markup included.
• Give reviewers specifics to quote; Google says unique, non-commodity content likely matters most in its AI guide.

Decision 4: Which one metric will you judge it by?

Use mention rate: the share of answers, across your fixed question list, in which an engine names your brand. It is the one common metric that measures a buyer hearing your name. Clicks are only a diagnostic: in Pew Research's March 2025 browsing data, people who saw a Google AI summary clicked a traditional result in 8% of visits versus 15% without one.

Position is worse than noisy. SparkToro's study, run with Gumshoe.ai, had 600 volunteers run 12 prompts through ChatGPT, Claude and Google AI Overviews or AI Mode 2,961 times. Two runs returned the same brand list less than once in 100 (Claude only slightly more often), and the same order about once in 1,000. Its advice: track visibility, the share of runs naming a brand, not position.

Presence is the usual problem, not sentiment: 144 of the 157 brand mentions in our audit records (self-selected brands, not a benchmark) were positive.

Define it tightly:

• One fixed list of questions, each asked in a few wordings.
• Each run more than once, on each engine at the default tier your buyers see.
• A mention counted only when the answer names you as an option to choose.

An "AEO audit" that skips these steps is a rank report with a new cover. Measuring AI visibility covers the method and tracker options.

What should you stop doing?

Stop paying for AI-specific hacks and keep the SEO fundamentals. Google's generative AI guide (July 2026) says optimizing for AI search is still SEO, and that for Google Search you can ignore:

• llms.txt as a lever. Google Search ignores it, and we know of no major engine that documents reading it for answers (our view).
• Writing for AI. Chunking pages into tiny pieces, rewriting them in a special style, or adding special schema.org markup.
• Inauthentic mentions. Seeking them is less helpful than it looks; earn real ones.

What do the first 30 days and the first 6 months look like?

Spend the first 30 days on what you control (questions, baseline, key pages, crawl access) and months 2 to 6 earning third-party proof and re-measuring. Judge the strategy at month 6, not day 30.

At month 6, keep the questions where mention rate moved and drop the rest. For a baseline without the manual work, our free audit asks ChatGPT three unbranded buyer questions and shows the full answers; the $29 report covers 10 questions on five engines.