80 brands · Frozen ±0.2% prices · Listings shuffled on each visit

Live, in-the-wild visibility evaluation

A benchmark for evaluating Generative Engine Optimization in real-world search environments.

LIVE-GEO measures whether GEO strategies can improve visibility across web-connected AI systems under controlled, reproducible conditions.

Browse the catalogueDownload JSONGitHub

80 brands · 480 listings · round 001 · October 2026 · next round: TBD

Product order changes on every visit; each listing keeps its frozen price.

AgenticGEOCOREC-SEOE-GEOStatistics Addition
A sapling with 5 GEO strategies. The tree is planned to grow larger once there are more than 15 GEO strategies, excluding Original.AgenticGEO · CORE · C-SEO · E-GEO · Statistics Addition

How it works

  1. Rewrite

    Download the raw JSON and rewrite only the descriptions and reviews, with your own method.

  2. Publish

    Submissions open with round 002. For now, publish your rewritten listings at your own URL and follow the Ask step using the same evaluation protocol.

  3. Ask

    Ask any web-connected AI for the best product in a category. Record the rank, the engine and the date. A listing the AI could not reach has no rank.

See the three prompt templates

These follow the three prompts of our 150-category study. Fill in the parts in braces: {category} is the category name and {N} is the number of listings in the category over all its brands (today one brand with its Original and five method versions, so 6). A family page is https://livegeobench.com/{family}/, a category page https://livegeobench.com/{family}/{category-id}/, a brand page https://livegeobench.com/{family}/{category-id}/{brand}/ and a listing https://livegeobench.com/{family}/{category-id}/{brand}/{listing}.html. The catalogue index (JSON) lists every URL.

1. Product URLs ChatGPT

For the query "best {category}", open the {N} product listings below and rank them from 1 (best) to {N}.

{listing URL 1}
{listing URL 2}
…
{listing URL N}

Return a Markdown table with columns:
Rank | Product name | Brand | Product URL | Read full description? | Read full comments? | Brief reason | The order in which items are fetched.

Include each listing exactly once. Report reading status honestly; use partial or no when appropriate. Base your reasons on the information you could access.

2. Category page ChatGPT

For the query "best {category}", start from the category page below. It lists the brands in this category. Open each brand's page to find its product listings, including each brand's Original, then open every listing yourself and rank all {N} listings from 1 (best) to {N}.

{category page URL}

Return a Markdown table with columns: Rank | Product name | Brand | Product URL | Read full description? | Read full comments? | Brief reason. Include each listing exactly once. Report reading status honestly; use partial or no when appropriate.

3. Category page Gemini

For the query "best {category}", start from the category page below. It lists the brands in this category. Open each brand's page to find its product listings, then open every listing yourself and rank all {N} listings from 1 (best) to {N}.
{category page URL}
Return a Markdown table with columns: Rank | Product name | Brand | Product URL | Read full description? | Read full comments? | Brief reason | the order in which items are fetched. Include each listing exactly once. Report reading status honestly; use partial or no when appropriate.

A listing the AI could not open is recorded as not observed, never as last place.

Catalogue

Every brand is fictional, and the names are deliberately unusual to reduce prior brand familiarity. Each listing has a fixed price within ±0.2% of its brand's base price.

18 families · 80 categories · 80 brands · 480 listings · Catalogue index (JSON)