AEO and GEO: the evidence

Which GEO and AEO tactics actually have evidence behind them?

Out of 80 tactics mentioned by enough publishers to compare, 30 are both recommended and backed by measurements, while 14 are openly contested and 12 are mostly called useless or risky. We built this table by saving 435 guides, studies, forum threads and official docs from 320 publishers and checking every quote word for word. Counts do not prove a tactic works, because many writers in this field sell GEO services or tracking tools.

Updated 19 November 2026

Across the 80 tactics with enough sources to compare, 30 are recommended and backed by published measurements. Another 24 are widely recommended but barely measured, 14 draw real pushback or failed tests, and 12 are mostly called useless or risky. So much of the standard AEO and GEO advice is repeated more often than it is tested. Even a tactic labelled measured usually rests on agency case studies or observational data, not controlled experiments.

How we label each tactic

On your own pages

StepRecommend itMeasured it workingCall it useless or warnMeasured no effectMeasured harmOur label
Answer first, in short self-contained passages an assistant can lift336020Recommended and measured
Question-and-answer pages matching the prompts people type174300Recommended and measured
Add statistics, quotations and cited sources143020Recommended and measured
Keep pages fresh and dated116010Recommended and measured
Comparison, best-of and alternatives pages on the business's own site206010Recommended and measured
Publish original data or research161010Recommended, little measured
Cover the sub-questions an assistant fans a prompt out into115011Recommended and measured
A page for each specific long-tail use case or prompt187000Recommended and measured
State plainly and consistently who the business is, what it does, where, for whom, prices364030Recommended and measured
Named expert authors, credentials, trust signals120030Recommended, little measured
Ordinary search ranking first (Google and Bing), since assistants search236650Contested
Bing indexing, Bing Webmaster Tools, IndexNow62100Recommended, little measured
Public pricing, help and policy pages143200Recommended and measured
Video with transcripts on the site43000Recommended and measured
Tables, lists and other structured formatting on the page40000Recommended, little measured
Plain, fluent, confident writing12000Recommended, little measured

On other sites

StepRecommend itMeasured it workingCall it useless or warnMeasured no effectMeasured harmOur label
Genuine participation on Reddit1910321Recommended and measured
Get included in third-party best-of lists and roundups156010Recommended and measured
Press, digital PR, brand mentions on trusted sites345200Recommended and measured
Review platforms (G2, Capterra, Trustpilot, Yelp, Tripadvisor)87120Recommended and measured
Wikipedia and Wikidata114010Recommended and measured
YouTube245000Recommended and measured
Quora, forums, LinkedIn posts, Medium, Substack143100Recommended and measured
Business listings and directories complete and consistent254220Recommended and measured
More, newer and more detailed customer reviews164110Recommended and measured
Affiliates, partners, publishers and marketplaces that already get cited21000Recommended, little measured
Publish your own list that ranks yourself66811Contested
Backlinks01020Mostly called useless or risky
Claim the business's profile inside the assistant (ChatGPT Business Hub and the like)50200Contested
Pay for a place on a list that already gets cited20300Mostly called useless or risky

Technical access for crawlers and agents

StepRecommend itMeasured it workingCall it useless or warnMeasured no effectMeasured harmOur label
Let the AI crawlers in (robots.txt)180001Recommended, little measured
Content in the HTML, not behind JavaScript172001Recommended, little measured
schema.org structured data242451Contested
llms.txt801180Mostly called useless or risky
Markdown or plain-text versions of pages for agents22110Contested
Product or service feeds200300Recommended, little measured
Clean, fast, semantic HTML62000Recommended, little measured
Make sure the CDN or firewall is not blocking agents101300Contested
Sitemaps30000Recommended, little measured
Forms and buttons a browser agent can operate90100Recommended, little measured
No CAPTCHA, login wall or forced account in the agent's path; guest checkout70201Contested

Agents that book and buy

StepRecommend itMeasured it workingCall it useless or warnMeasured no effectMeasured harmOur label
An MCP server for the business143000Recommended and measured
An app or connector in the ChatGPT directory60410Contested
A connector in the Claude directory43300Contested
Agentic Commerce Protocol and Instant Checkout (OpenAI, Stripe)10522Mostly called useless or risky
Google's routes (Universal Commerce Protocol, Agent Payments Protocol, Reserve with Google, Business Agent)70100Recommended, little measured
Be on a platform that is already wired in203210Recommended and measured
A public, documented API for booking or ordering40100Recommended, little measured
Agent payment rails80230Contested
Agent identity and trust (Web Bot Auth, signed agents)50000Recommended, little measured
Agent-to-agent discovery (A2A agent cards, NLWeb, registries)20200Contested
Live price, stock and availability in machine-readable form161100Recommended, little measured
Write the directory listing, tool names and tool descriptions so the model picks the tool50100Recommended, little measured
WebMCP: page actions exposed as tools to a browser agent40000Recommended, little measured
A business agent that answers a customer's agent30000Recommended, little measured

Measuring

StepRecommend itMeasured it workingCall it useless or warnMeasured no effectMeasured harmOur label
A fixed panel of prompts run on a schedule; share of answers naming the business5441100Recommended and measured
Analytics set up to see AI referrals3511000Contested
Ask every new customer how they found the business194201Recommended and measured
Server logs for AI crawler and agent visits60100Recommended, little measured
Find which sources the assistants cite for the target prompts, then get onto those4110400Recommended and measured
Run each prompt many times, since answers vary256510Recommended and measured
Track how AI visitors convert against other visitors268510Recommended and measured
Find the real prompts customers use183100Recommended and measured
What AI visibility is worth (stated value, not a method)11400Mostly called useless or risky
Sell and price the service on measured outcomes20420Mostly called useless or risky

Tricks

StepRecommend itMeasured it workingCall it useless or warnMeasured no effectMeasured harmOur label
Hidden text or prompt injection aimed at the model12910Mostly called useless or risky
Mass-produced AI pages001200Mostly called useless or risky
Fake reviews or planted Reddit posts21710Mostly called useless or risky
Expired domains, parasite pages, paid link schemes01800Mostly called useless or risky

How an agent picks from a list

StepRecommend itMeasured it workingCall it useless or warnMeasured no effectMeasured harmOur label
Position in the list the agent reads01021Mostly called useless or risky
Rating and review count shown to the agent13000Recommended and measured
Price against the other options; same price for agent and human12000Recommended, little measured
Badges and labels on the listing (platform pick, sponsored)11001Contested
Title and description wording: plain facts, the searched words, cues that match the request32011Recommended, little measured
A third-party pick (review site, forum) the agent reads before it shops12000Recommended, little measured
Each model weighs the same factors differently03000Recommended and measured
The listing or booking platform blocks agents, so its businesses are unreachable10200Mostly called useless or risky
Persuasion text, fabricated authority and social-proof claims02020Contested
The business's own shopping or booking assistant on its site03000Recommended and measured
How often agents complete a purchase today40000Recommended, little measured

We saved 435 guides, studies, forum threads and platform docs from 320 publishers in full, recorded 1592 word-for-word method statements across 99 tactics, and logged 216 reported results with numbers. Every count is by publisher, not by article, so one prolific blog only gets one vote per stance, and every quote is checked word for word against our saved copy. A high count is not proof that a tactic works: many GEO writers sell GEO services or software, and a measured claim usually means one site's before-and-after numbers rather than a controlled test.

Common questions

Does a high publisher count prove a GEO tactic works?

No, a high count only shows what writers in the field agree on right now. Many people publishing AEO and GEO advice sell GEO services or tracking software, and many repeat the same tips without testing them. Look at the measured columns and read the word-for-word quotes to see whether anyone ran a controlled test or just watched one site's traffic.