Your CMO asked you to “Get us cited by ChatGPT”. Here is what to do next.
Analysis / AI search visibility
AI Search Visibility Audit: 50 Checks in Four Phases
Most companies are invisible to AI search for reasons that take an afternoon to fix. They stay invisible because nobody ever wrote down what the work contains. This is that list: 50 checks, four phases, one score out of 100.
The cheapest reason to be absent from ChatGPT answers is a single line in your robots.txt. A security plugin added it. Nobody approved it. It takes fifteen minutes to find and remove.
That fix rarely happens, and the reason is not laziness. It is that AI search visibility arrives as a vague instruction rather than a scoped project. Someone says get us cited by ChatGPT, and the work has no boundary, no owner and no first task. So the team debates a replatform instead, because a replatform at least has a shape.
The ask sounds like a task. It behaves like a program.
Kaleigh Moore made the sharpest version of this point in September 2026, and it is worth stating plainly: getting cited by ChatGPT is not one job. It is standing work in three directions at once.
- Measurement never finishes. An answer is probabilistic, not a fixed rank, so one screenshot proves nothing. You run the same questions in clean sessions, month after month, and the tools tell you your presence rate without telling you why.
- The question set has to be built and maintained. It comes from what buyers actually say on calls and in tickets, not from an executive brainstorm, and it goes stale as the category moves.
- Most of the work is not on your website. The pages an answer engine quotes about you sit on review sites, forums, roundups and other people’s platforms. That means coordinating people who do not report to you.
Accept all of that and you still cannot start on Monday. Knowing how big the work is does not tell you what to change first.
That is the gap. A new Head of AEO gets a baseline, a question set and a board meeting in six weeks. What nobody hands them is a scope document: a finite list of changes, each one testable, each one with an honest cost against it.
Free tool
AI Search Visibility Audit
Fifty checks in four phases. Mark each one, get a score out of 100, and get a ranked list of what to do next with the steps for every fix. Runs in your browser. No account needed to use it.
Run the auditFour phases, ordered by what they cost you
The audit sorts the work by lift, not by topic. That ordering matters more than the checks themselves, because it decides what you can start without asking anyone.
The four phases
Each phase has a different owner and a different approval path. Read the owner column as the conversation you have to win.
Settings, files and profiles. No content changes. One person, one working day.
Edits to pages you already own, plus the off-site work that earns citations.
New systems, not new pages. Content, web and analytics have to move together.
Architecture and org design. Needs a budget line and an executive who defends it.
| Phase | Owner | Example check | Checks | Points |
|---|---|---|---|---|
| 1. Quick wins | One person | Unblock AI crawlers in robots.txt | 13 | 26 |
| 2. Content work | Content and PR | Publish comparison and alternative pages | 14 | 28 |
| 3. Structural work | Content, web, analytics | Rewrite key pages as answer first passages | 12 | 24 |
| 4. Rebuild | Leadership | Expose a live query interface such as MCP | 11 | 22 |
Every check is worth two points. Fifty checks make a score out of 100.
More than half the score is cheap
Here is the number that should change your plan. Twenty-seven of the fifty checks sit in the two cheapest phases. That is 54 of the 100 points available for work measured in hours and weeks.
Points by phase
Phases 1 and 2 together carry more of the score than phases 3 and 4 combined.
Bars show each phase as a share of the 100 point total.
Most teams work the list backwards. They open the conversation at phase 4, where the ideas are interesting and the timelines are long, while phase 1 sits untouched. A knowledge graph does not help a company whose robots.txt blocks the crawler that would read it.
Where the fifty checks come from
Forty of them start from a list published by 1000startups in July 2026. Its useful move is the sort order: it separates what you can do today without touching your content from what needs real change, and both from what needs the business rebuilt. The audit keeps those forty items, rewrites each one as a check you can pass or fail, and reorders them by lift.
Ten more checks close gaps the original list leaves open.
The ten added checks
Each one covers something the original forty does not reach.
| Added check | Phase | The gap it closes |
|---|---|---|
| No cookie wall or login hides your content | 1 | A consent banner or paywall can serve a crawler the wall instead of the page. |
| Fix crawl errors and broken links | 1 | Dead links and redirect loops spend the crawl on nothing. |
| Correct the wrong facts AI answers give about you | 1 | Tracking finds the error. Nothing says to trace it to its source page and fix it there. |
| Publish comparison and alternative pages | 2 | Pitching your way onto other people’s roundups is covered. Owning the versus question yourself is not. |
| Publish your prices | 2 | A price behind a form cannot be quoted, so engines answer with whoever published a number. |
| Format your answers for extraction | 2 | Answer-first prose is covered. Tables, numbered steps and standalone definitions are not. |
| Write alt text and publish transcripts | 2 | Images and video are absent from all forty, and both are cited surfaces that need words. |
| Fix crawl hygiene across the whole site | 3 | Core Web Vitals, canonicals and redirect chains at scale, not clean HTML on one template. |
| Open your documentation and support answers | 3 | Your support team writes the best answers you have. Behind a login, none can be read. |
| Publish a product feed to the agentic commerce protocols | 4 | Assistants complete purchases in the conversation now. Inclusion runs on a feed and checkout endpoints. |
The comparison page and crawl hygiene checks follow an AEO checklist from Stackmatix. The product feed check follows a 2026 guide to agentic commerce protocols from Opascope, which covers the Agentic Commerce Protocol used by ChatGPT and Google’s Universal Commerce Protocol. That area moves fast, so read the current merchant documentation before you scope it. Both links are in the sources below.
Turn the score into a budget request
A score out of 100 is a talking point. A score split by phase, with a named owner and an honest cost against each one, is a budget request. This is the part that matters if you own AI visibility and have to defend the resourcing.
| Phase | What you are really asking for | How to open it |
|---|---|---|
| 1 | Nothing. One person for a week. | Do not ask. Clear it before the first review so the meeting starts from a better number. |
| 2 | Content and PR capacity, permanently. | Six of its checks are marked ongoing, because reviews, forums and a PR programme do not finish. Ask for the standing capacity, not a project. |
| 3 | Three teams moving together for a quarter. | Needs a sponsor rather than a volunteer. Bring the phase 1 and 2 gains as evidence. |
| 4 | A budget line and an executive who defends it for a year. | Only after phases 1 to 3 are behind you. Earlier than that, it reads as a wish. |
If you only get one hire. Fund the measurement job first. Checks 09 and 13 in phase 1 set the baseline and correct what the engines already get wrong about you. Check 36 in phase 3 makes AI referral traffic visible in analytics, where most of it currently lands under Direct with no referrer. Without those three you cannot prove that anything you do next worked, and the budget for phases 3 and 4 never arrives.
How to run it
- Answer every check honestly. No, Partly and Yes all count as answers. About twelve minutes.
- Open the question mark on anything you are unsure about. Each check has a panel with why it matters, five steps, a test for done, and the mistake most teams make.
- Read the ranked list at the end. It sorts by points earned per day of work, so the cheap wins rise to the top.
- Download the scorecard. It carries your score, every answer and the steps for the eight highest-return fixes.
- Re-run it each quarter. Visibility decays when content goes stale, so a score you earned once does not stay earned.
Common questions
What is answer engine optimization?
Answer engine optimization is the work of getting your business found, read and quoted by AI answer engines such as ChatGPT, Perplexity, Gemini and Claude. It overlaps with SEO on the technical side. It differs on the content side, because the goal is to be quoted inside an answer rather than to rank in a list of links.
How is the audit scored?
There are fifty checks. Each is worth two points, so the maximum is 100. Yes scores the full two points, Partly scores one, and No scores zero. The score maps to five bands, from Invisible at the bottom to Cited by default at 80 and above.
What should a team do first?
Phase 1. Thirteen of the fifty checks are settings, files or profiles rather than content, and one person can clear them in a working day. The cheapest takes fifteen minutes and is the most common reason a company is invisible: a robots.txt rule blocking AI crawlers, usually added by a security plugin that nobody remembers approving.
Do I need a paid tool to measure AI visibility?
Not to start. Pick ten to twenty buyer questions in the words a buyer would use. Run them monthly in a fresh session with memory and personalization switched off. Record four things each time: were you named, were you linked, who else was named, and which pages were cited. Paid tools do this at larger scale. None of them explain why a citation happened.
Why does the audit order phases by lift instead of by impact?
Because lift is the part you can verify and impact is the part you cannot. Effort is knowable before you start. A claim that one tactic returns more than another depends on your category, your competitors and which engine you care about. Ordering by lift lets you decide what to do this week without trusting a ranking that nobody can prove.
How often should the audit be re-run?
Each quarter. Answer engines favour recent content, and several checks are ongoing rather than one-time. A freshness calendar with a named owner is one of the fifty, and it exists because a one-time cleanup decays quietly.
Sources and further reading
Just get us cited by ChatGPT
Kaleigh Moore, 9 September 2026. Why the ask is standing work rather than a task, and why measurement, question sets and off-site content each need their own time.
Read the piece40 things to do to make your website more visible to AI search
1000startups, 23 July 2026. The forty item list the audit builds on, sorted by difficulty.
Open the listAEO optimization checklist
Stackmatix. Source for the comparison page and crawl hygiene checks added to the audit.
Open the checklistAI shopping assistant guide 2026: agentic commerce protocols
Opascope. Source for the product feed check and the two protocol names.
Open the guideEffort estimates in the audit are planning judgements, not measured figures. Every one is labelled as an estimate inside the tool.
