GEO checklist, graded by evidence

A GEO checklist where every item says how strong the evidence is: platform documentation, a published study, or untested. 19 checks with sources and a free Excel download.

Last updated

On this page
  1. The grades
  2. 1. Measure first
  3. 2. Do next: documented or studied
  4. 3. Skip
  5. 4. Optional: untested, so test before you rely on it
  6. 5. Never
  7. Why grade a GEO checklist?
  8. Sources and corrections
  9. Related
  10. License and credit
  11. Update log

Short answer: a generative engine optimization (GEO) checklist where every item says how strong the evidence behind it is. Start with three measurement steps, then the six items that Google, OpenAI, IndexNow or a published study support. Skip special AI files for Google: Google says you do not need them. Treat the eight untested items as optional experiments, and remove one practice entirely. Free Excel version below, with a status column and a progress tab.

Free checklist · Excel · CC BY 4.0

GEO checklist, graded by evidence

19 checks with sources, a status column you fill in, and a progress tab by grade. No signup.

Download (.xlsx)

The grades

GradeMeaning
[T]Tested by the AI Visibility Census, with the result published
[S]Published study with a stated method
[P]Platform documentation: the search engine or AI company says so in its own docs
[V]Vendor or practitioner claim, not independently checked
[U]Untested, or no evidence found either way
MethodHow to measure, not a ranking tactic
Do not useHarmful or deceptive

A higher grade is not a promise. [P] tells you what a platform says it does, not what happens on your site. [S] results come from the study’s setup, which may not match yours. No item is graded [T] yet: the Census experiments in the last column have not reported results.

1. Measure first

Check and how to check itGradeWhat the evidence saysCensus test
Write a fixed set of buyer questions before you change anything
List the questions buyers ask, spread across vendor selection, comparison, specific need and educational intents. Keep the wording fixed so months compare.
MethodThis is how you measure, not a ranking tactic. Without a fixed question set, a change in results cannot be told apart from a change in the questions. Question set builder (free skill)Used for Census Edition 1 and CensusRank™
Ask each question at least 3 times per engine per check date
Log every answer with the date, engine, run number, answer text and cited links.
MethodAI answers change from run to run. With 1 or 2 runs the only possible rates are 0%, 50% or 100%. Answer log analyzer methodsEdition 1 plan: 3 runs per date on ChatGPT, Gemini and AI Mode
Change one thing at a time, with a control page and a written prediction
Note the date of each change and what you expect before results come in.
MethodEngine updates land between check dates too, so a change you cannot isolate tells you little. GEO experiment kit (free skill)Every test on our experiments page

2. Do next: documented or studied

Check and how to check itGradeWhat the evidence saysCensus test
Pages you want in Google's AI features are indexed and eligible for a snippet
URL Inspection in Search Console shows the page is on Google. No nosnippet on the page.
[P]Google: pages need to be indexed and eligible to be shown with a snippet; there are “no additional requirements”. Google, AI features and your websiteNot tested
OAI-SearchBot is allowed in robots.txt
Open yoursite.com/robots.txt and check nothing disallows OAI-SearchBot (or every bot with *).
[P]OpenAI: sites opted out of OAI-SearchBot “will not be shown in ChatGPT search answers”. GPTBot is the separate training crawler. OpenAI, botsNot tested
Important content is in the HTML the server sends, not only added by JavaScript
View the page source (not the inspector) and search for a key sentence.
[S]Vercel and MERJ (December 2024): “none of the major AI crawlers currently render JavaScript”; exceptions were Gemini (via Googlebot) and Applebot. Data from Vercel’s network, checked on two other sites, 2024. Vercel, The rise of the AI crawler
Census data so far: on 4 November 2026, OpenAI’s GPTBot made 156 verified requests to our site in 105 seconds, and 18 were for raw pieces of our JavaScript code requested as links: it read the script as text. That does not show whether it runs JavaScript, so the grade stays [S]. Source: our article What is GPTBot?
Not tested
IndexNow sends new and changed URLs to Bing
Check your CMS or plugin sends IndexNow pings and gets a 200 or 202 back.
[P] partialIndexNow lists Bing among participating engines and says “Submitting a URL does not guarantee immediate indexing”. The speed claim is not documented. IndexNow FAQUsed within CT-001
Structured data says only what the visible page says
Compare schema FAQ text, prices, ratings, phone numbers and dates against the visible page.
[P]Google lists making sure structured data matches the visible text among practices for AI features. Google, AI features and your websiteNot tested
Add statistics, quotations and citations to key pages
Each key page carries at least one number with its source, and names its sources.
[S]Aggarwal et al.: “GEO can boost visibility by up to 40% in generative engine responses”, in a benchmark setting. Lab results, not a live-engine guarantee. arXiv 2311.09735Not tested

3. Skip

Check and how to check itGradeWhat the evidence saysCensus test
Do not add special AI files or markup expecting it to get you into Google's AI features
Nothing to add. If a vendor sells you an AI file for Google, ask for the source.
[P] againstGoogle: “You don’t need to create new machine readable files, AI text files, or markup to appear in these features”. Google, AI features and your websiteNot tested

4. Optional: untested, so test before you rely on it

Check and how to check itGradeWhat the evidence saysCensus test
FAQ structured data
Optional. Keep it only if it matches visible text.
[U]Google stopped showing FAQ rich results from 7 May 2026, so there is no Google display benefit. No evidence found that it helps AI answers. Google Search Central updatesIn place within CT-001; not isolated
llms.txt
Optional. Cheap to add, but do not expect results from it alone.
[U]A proposal by Jeremy Howard, first published September 2024. Google says AI text files are not needed for its AI features. Whether AI crawlers fetch it is what the Census is logging. llmstxt.org
Census data so far: in our first three days, verified Googlebot, GoogleOther and ClaudeBot requested llms.txt or llms-full.txt 10 times out of 440 verified crawler requests (about 2%); Bingbot and Applebot never asked. A fetch is not an effect, so the grade stays [U]. Source: our Lab note 002, 29 October to 1 November 2026.
LL-001 running; RT-001 from 1 Dec 2026
Markdown versions of pages
Optional.
[U]No platform documentation or study found.
Census data so far: in the same three days, the same verified crawlers fetched Markdown copies of our pages 77 times, about 17.5% of all 440 verified crawler requests and about eight times their llms file requests. Fetches, not citations, so the grade stays [U]. Source: our Lab note 002, 29 October to 1 November 2026.
Logged within CT-001
Content Signals line in robots.txt
Optional. Know that no AI company has been found committing to honor it.
[U]A robots.txt extension introduced by Cloudflare in September 2025. No AI company found committing to honor it. Cloudflare, Content Signals PolicyCS-001 running
Question-style headings
Optional.
[U]Claimed to get quoted word for word. Not checked here.Not tested
Rank in Google's top 10 for the question
Check your ranking for the main query behind each question.
[U]Claimed to be required for AI Overview citations. Claims vary between vendor studies; not checked here.Not tested
Earn backlinks to the pages you want cited
Referring domains to each key page, from Ahrefs, Semrush or Bing Webmaster Tools.
[U]Claimed that a page with no backlinks will not be cited. Not checked here.Not tested (BL-001 measures links, not citations)
Share new pages on LinkedIn to speed up the first Bing crawl
Optional.
[U]Claimed to speed up the first Bing and Copilot crawl. Not checked here.Not tested

5. Never

Check and how to check itGradeWhat the evidence saysCensus test
Hidden instructions aimed at AI assistants
Check “Summarize with AI” buttons and share links on your site: none should carry a prompt telling the assistant to remember or favor your brand.
Do not useMicrosoft describes companies hiding instructions that tell an assistant to remember and favor their brand, and calls it “AI Recommendation Poisoning”. Microsoft Security, 10 February 2026Never

Why grade a GEO checklist?

A checklist that lists llms.txt next to letting OpenAI’s search crawler into your site treats them as equal. They are not. Google’s documentation says “You don’t need to create new machine readable files, AI text files, or markup to appear in these features” (Google, AI features and your website). OpenAI says sites opted out of OAI-SearchBot “will not be shown in ChatGPT search answers” (OpenAI, bots). One is documented by the platform; llms.txt is a proposal, and Google says AI text files are not needed for its AI features. Grading them lets you spend time where the evidence is.

When an untested item gets tested, by us or by a published study, its grade moves. Census tests and results go on the experiments page, whatever they show.

Sources and corrections

Grades and sources come from the geo-skills evidence file, whose sources were last checked 29 October 2026. If a grade is wrong or a source has changed, open an issue with the link or email hello@aivisibilitycensus.com. Grades move up only when the evidence does.

License and credit

Free to use and adapt under CC BY 4.0, including for client work, with credit and a link: AI Visibility Census, “GEO checklist, graded by evidence”, aivisibilitycensus.com/templates/geo-checklist/. The file that builds the sheet is on GitHub, with a Markdown copy of this checklist. Version 1 is archived on Zenodo with a permanent DOI: 10.5281/zenodo.23960475.

New templates and every Census test result, sent when they publish: get the Census by email. Free, and one click to leave.

Update log

DateChange
4 November 2026Checklist published with 19 items. Evidence sources as checked on 29 October 2026.
4 November 2026Version 1 archived on Zenodo (DOI 10.5281/zenodo.23960475).
5 November 2026Census crawler data from Lab note 002 added to the llms.txt and Markdown rows. Grades unchanged. The Excel file and the Zenodo record (version 1) are unchanged.
11 November 2026Census crawler data on GPTBot and JavaScript added to the server HTML row, with a link to our GPTBot article. Grade unchanged. The Excel file and the Zenodo record (version 1) are unchanged.
12 November 2026The Census100 is now part of CensusRank™; one name changed in the Measure first table. Grades, the Excel file and the Zenodo record (version 1) are unchanged.