Methodology
What the audit checks, and why.
CiteGap is the AI visibility audit by Demand Field Notes. This page documents what it measures, how the scoring works, and what a run looks like from start to finish.
Six checks, from "can they read you" to "would a buyer trust you".
Discovery
Can AI engines reach your site at all? If a crawler is blocked or a page is missing, nothing else on this list matters. We check what engines are allowed to fetch and whether your key pages resolve cleanly.
Why: engines cannot cite what they cannot fetch.
Readability
Once a page is fetched, can an engine parse and quote it? We check whether your pages lead with a plain answer, carry clean structure, and expose facts machines can lift without guessing.
Why: answers are assembled from quotable text.
Recommendation
When a buyer asks a comparison question, are you in the answer? We log, per question and per engine, whether you are named, hedged, or left out entirely, and where you appear in the order.
Why: the comparison question is where vendors get picked.
Consistency
Does what AI says about you match what your own site says? Prices, category labels, and company details drift as engines hold older snapshots. Conflicts between versions make engines hedge.
Why: an engine that cannot pin down your facts stops short of recommending you.
Contactability
When you are named, does the answer point somewhere the buyer can act on? We check whether your official site is the destination, or whether an aggregator or directory intercepts the visit.
Why: being named through someone else's page is a lead you never see.
Decision Evidence
Close to signing, buyers ask engines harder questions: is this vendor trustworthy, what results have they delivered, what are the risks. We test those questions and check whether citable proof about you exists at that moment.
Why: no citable evidence means vague praise, and vague praise does not close deals.
What the report scores.
The paid audit logs 30 buyer-style questions across ChatGPT, Perplexity, and Gemini, run twice over two days. For every question and every engine we record the same facts, so scores come out of the logs instead of a gut feel:
Named or not. Whether the answer names your product, a competitor, or nobody.
Cited or not. Whether the answer cites sources, and whether any citation points at your site.
Order. Whether you come first, mid-list, or as an afterthought.
Hedging. Whether the engine recommends you outright or qualifies the mention into uselessness.
Day-over-day movement. What changed between the two runs, per engine.
Gap list. Where you lose, with the question and engine that shows it.
Every gap lands in the fix list as a five-field fix ticket, ordered by priority, so you can hand the list to whoever ships the changes. The report, the raw answers, and the fix list are yours to keep, delivered within five business days of payment.
The free instant check is the fast version: it examines seven technical aspects of your homepage immediately and shows the result of each. It covers the Discovery and Readability layer only. No signup required. Run it here.
The reasoning behind the method.
Questions written like a buyer writes them
Ranking checks ask a database. Buyers ask an engine a sentence: "which tool should we use for X, and what should we check before signing". The audit's questions use that phrasing, with your product, category, and competitors substituted in. Here is what that looks like in one industry.
Comparison questions get answers
When someone asks an engine to compare vendors, the engine has to commit to names. That is the moment your absence costs you, and it is the moment the audit measures hardest. Being named beats ranking.
Measured twice, because answers move
The same question can return a different vendor tomorrow. A single run would make that movement invisible, so every run happens on two days, and the difference between them is part of the read. Why scores disagree between surfaces.
From first question to fix list.
The first run
The question set goes to ChatGPT, Perplexity, and Gemini, one question at a time, exactly as a buyer would type it. Every answer is logged: who was named, in what order, and which sources were cited.
The re-run
The same questions run again on a fresh day. Names move between days; capturing both runs keeps the report honest about what is stable and what is luck.
Scoring
The two runs are read together: named-rate per engine, citation sources, and the specific questions where you lose to a competitor or vanish entirely. That becomes the gap list.
The fix list
Every gap becomes a five-field fix ticket, ordered by priority: what to change first, and what each change is expected to unblock. You receive the report, the raw answers, and the fix list, yours to keep.
What we have put in writing so far.
The method above is not a pitch deck artifact. It is backed by published work you can read before spending anything: 6 industry briefs and 14 field notes, each with its sources named.
AI Visibility for DevTools
Is ChatGPT recommending your tool
AI Visibility for Law Firms
Is ChatGPT recommending your firm
AI Visibility for Healthcare
How AI describes your clinic
AI Visibility for Dental Clinics
Is ChatGPT recommending your practice
AI Visibility for Local Businesses
Does AI recommend your business
AI Visibility for B2B SaaS
Are you on the AI shortlist
The full archive of field notes lives in the Notes Archive.
A lab of one, run in the open.
Demand Field Notes is an independent AI intelligence lab. One practitioner runs it: the person who writes the questions runs the tests, reads the answers, and writes the report. There is no dashboard account to maintain and no sales team to route around. When you reply to your report, you reach the operator.
We hold the audit to the standard we sell. The first full CiteGap run is on our own site, demandfieldnotes.com, with the same question set and the same two-day measurement we run for clients. The results are documented as the first case study, including what we found wrong on our own pages.
Nothing on this site is ranked, rated, or sponsored. When we cite a study, we link it and name its sample size. When we do not know something, the report says so.
Terms, refunds, and your data. Refund terms are in the terms of service. What gets stored during an audit, and how to have it deleted, is in the privacy policy. Deletion requests go to hello@demandfieldnotes.com.
Start with the free check.
Seven technical aspects of your homepage, checked instantly, no signup. The full 30-question audit across three engines can follow.
Questions about the method first? hello@demandfieldnotes.com