AI shortlist test (GEO)
Checks whether ChatGPT, Perplexity and Gemini name you when a buyer asks for suppliers: turns your ICP into the buyer's questions and reads the answers
Phase 3 · Demand generation engine › SEO and GEO
SEO and GEO
Buyers ask AI assistants which suppliers to consider, and most teams don't know whether they're on the list.
Goal
Be found on Google and named in AI answers for the questions buyers ask when they choose a supplier.
Key metric
Share of the buyer's questions where search and AI answers name you; pipeline from organic search and AI assistants.
Common mistake
Chasing keywords with high volume and far from the purchase, or treating GEO as a technical trick when assistants mostly cite other people's sites.
Who owns it
Content and SEO, with product marketing and PR; marketing ops for measurement
What changes
SEO gets your pages found by search engines. GEO, generative engine optimisation, gets you named in the answers of AI assistants such as ChatGPT, Perplexity and Gemini. More and more B2B buyers ask an assistant which suppliers to consider before they visit any website or talk to sales, so the shortlist starts there.
| SEO · search engines | GEO · AI answers | |
|---|---|---|
| What the buyer types | A few keywords | A full question, often with their context: company size, industry, the tools they already use |
| What they get back | A page of links to choose from | One answer that names a few suppliers, with or without links |
| What wins | Your page ranks high | You're named, and the sources the assistant relies on mention you |
| Where you work | Mostly your own pages | Your pages, plus review sites, communities, comparisons, partners and the press |
| How you measure | Clicks and position | Share of the buyer's questions where you're named, plus what buyers tell you |
SEO still matters: assistants that search the web start from pages that search engines can find and read. GEO adds a layer on top, it doesn't replace SEO.
SEO still matters: assistants that search the web start from pages that search engines can find and read. GEO adds a layer on top, it doesn't replace SEO.
What to do
Answer the buyer's questions
One page for each question that matters in the purchase, written in the buyer's words. The answer comes in the first lines, the detail after. Start from the questions sales hear on calls, not from keyword volumes.
Publish facts others can cite
Your own data, price ranges, who the offer is for and who it isn't for, and comparisons that admit where a competitor is the better fit. Assistants repeat specific, checkable facts; they skip slogans.
Be present where assistants look
Review sites, communities, partner pages, industry articles and analyst reports. You can't write these yourself, but you can earn them: ask customers for reviews, take part in discussions, give journalists data.
Make your pages readable by machines
Text in the HTML and not only loaded by JavaScript, clear headings, an author and a date, structured data, and crawlers allowed in. Many AI crawlers don't run JavaScript: what they can't read, they can't cite.
Where AI fits in
A typical case: the step in this topic where AI changes the work most, and who does what.
The monthly AI shortlist test
Generative AI- 1 · QuestionsAIClaude, with the skillTurns the ICP and buying jobs into 20 to 30 questions a buyer would askPeopleProduct marketingCuts the questions nobody asks and adds the ones heard on calls
- 2 · AskToolsChatGPT, Perplexity, GeminiEach question asked three times, in a new chat with no historyPeopleMarketing opsPastes the answers into one file
- 3 · ReadAIClaude, with the skillFor each answer: are you named, where, next to whom, with which sources, and is it accurateToolsGoogle Search ConsoleClicks and position on Google for the same topics
- 4 · CompareAIClaudeCompares with last month and lists the sources cited most where you're missing
- 5 · DecidePeopleContent and PRChoose the pages to write or fix and the outside sites to work on
BeforeNobody knows whether AI assistants mention you; sales find out by chance from a prospect.
AfterTwo hours a month: where you're missing, which sources count, and what to fix first.
The limitAnswers change between runs and between users: it's a sample, not a ranking. The AI can misread an answer, so a person checks the summary.
Indicative times, based on a typical case.
How to measure it
- A sample of questions. The same 20 to 30 buyer questions every month, on three assistants: are you named, next to whom, and with which sources. The AI flow above, with the skill below.
- Traffic from assistants. A separate source in your analytics for visits from chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Many visits arrive with no referrer at all, so this number is always too low.
- What buyers say. A free-text "How did you hear about us?" field on demo and contact forms, with "AI assistant" as a category when you classify the answers. More on this in phase 5.
Nobody can see how often an assistant names you across all its users. Together, these three signals show a trend, not a total.
Resources
SEO and GEO page checklist
Ten checks before publishing a page, so search engines and AI assistants can read and cite it
How to install the skill: in Claude, upload the zip in Settings, under Skills. In Claude Code, unzip the folder into ~/.claude/skills/. Then just ask: "Here's our ICP and three competitors. Please run the AI shortlist test."
- The main text is in the HTML, not only loaded by JavaScript · check: view the page source and look for your text
- The page answers one buyer question, and the answer is in the first lines
- The title and description say what the page answers, in the buyer's words
- Headings follow the questions, one idea per section
- Facts have a number, a source and a date · for example "updated October 2026"
- It says who the offer is for, and who it isn't for
- There's an author, linked to a real profile
- Structured data describes the page and the author · check: Google's Rich Results Test
- Neither robots.txt nor the CDN's bot settings block search or AI crawlers
- The page is in the sitemap and linked from other pages on the site
From the field · this site
I applied the checklist above to this site on 17 October 2026, the day it went online. The phase pages were built with JavaScript: a crawler that doesn't run it saw about 220 words of labels on the phase 5 page, and none of the method.
BeforeAbout 220 words readable without JavaScript on the phase 5 page. No sitemap, no structured data, no preview image for sharing.
ChangeThe text is written into the HTML when the site is generated (about 800 words on the same page), plus a sitemap, structured data and a robots.txt that welcomes AI crawlers.
NextThe AI shortlist test on this site's own topics, once a month from October 2026. The results will appear here, including if nothing changes.
Building AI into your demand gen?
Share what works for you, or ask about a resource.