The AI Search Audit Standard: 10 Checks Any Audit Should Pass
The AI Search Audit Standard is a public list of 10 checks that any AI search audit should pass before a buyer trusts its findings. Citant.ai scores its own $440 AI Search Audit against every check below.
You bought an audit, or you are about to, and the report says your brand is missing from ChatGPT answers. You cannot tell whether that verdict came from many careful runs or from one screenshot. Citant.ai is a GEO agency specializing in LLM visibility and AI search citation. It wrote these checks so you can test any audit, including its own, without trusting the seller.
A standard, in this sense, is a set of written checks that decide whether an audit’s numbers can be reproduced by someone other than the person who sold it. The checks cover coverage, evidence and terms. None of them asks for the seller’s private method.
First version, published 31 October 2026. Changes to the checks will be listed on this page with their date. Updated 2 November 2026: each check is now ranked must or should, and the AI Search Audit’s check 7 moved from FAIL to PARTIAL after a re-test before delivery was added.
Quick answer
The AI Search Audit Standard has 10 checks, and six are must checks. Five cover counting: named AI apps, results as a count with a base, naming apart from linking, a disclosed question list and repeat runs. Five cover evidence and terms: raw dated answers, a re-test, price before payment, an independent measurer and a stated remedy. An audit with no raw answers behind its findings gives its buyer nothing to check.
What does the AI Search Audit Standard check?
The AI Search Audit Standard checks three things: what the audit covered, what evidence backs each finding, and what the buyer was told before paying. Checks 1 to 5 cover coverage and counting, checks 6 and 7 cover evidence, and checks 8 to 10 cover terms. Each check states what passes, what fails, and where the AI Search Audit sold on this site stands, as stated in its published terms on 2 November 2026.
Which checks are must checks?
The must checks are 1, 2, 4, 5, 6 and 8: every AI app named, results as X of N, the question list disclosed, more than one run per question, raw dated evidence, and price, time window and deliverables stated before payment. An audit that fails a must check leaves the buyer unable to check its findings or to hold the seller to its terms.
The AI Search Audit Standard ranks each check as must or should, so a buyer knows which failures end the deal. A failed must check is a deal-breaker, and a must check that is only partly passed is a question to put to the seller in writing before paying.
Why are checks 3, 7, 9 and 10 ranked lower?
The should checks are 3, 7, 9 and 10: naming kept apart from linking, a re-test before delivery, an independent measurer and a stated remedy. An audit that fails one can still be checked, but its findings are easier to misread or the buyer has less recourse. The AI Search Audit sold on this site partly passes check 7 and fails check 9, and both stay should checks.
A re-test ranks below the must checks because check 5 already asks for repeat runs, so a re-test adds a second sample on top of them.
Check 1: is every AI app named?
An audit passes check 1, a must check, when the report names each AI app it tested, for example ChatGPT, Perplexity AI, Google Gemini, Claude, Microsoft Copilot, Google AI Overviews. An audit fails when it says “every AI engine” or “AI search” without a list. Each app retrieves and answers in its own way, so a result on one app cannot stand in for another.
The AI Search Audit: PASS. All 6 apps are named on the AI Search Audit service page.
Check 2: is every result a count with its base?
An audit passes check 2, a must check, when each result reads as X of N, for example a brand named in 3 questions out of 20, in 2 runs out of 3, on 4 apps out of 6. An audit fails when it prints a percentage, a grade or a score with no base. A share without a base cannot be checked or repeated.
The AI Search Audit: PARTIAL. Share of Model is reported per platform as a percentage of the tracked target queries, and the AI Search Audit service page states the base before payment: 20 buyer questions, 3 runs each, on 6 AI platforms. How the 3 runs of one question add up to one result is not stated.
Check 3: is naming kept apart from linking?
An audit passes check 3, a should check, when it counts two outcomes separately: the answer names the brand in its text, and the answer lists a brand page as a source. An audit fails when it reports one blended count.
Naming and linking diverge. Seer Interactive reported a citation rate of 53.1% when the brand is named in the response, versus 10.6% when it is not (Seer Interactive, March 2026). That study excludes Claude and Meta and describes behavior, not a proven mechanism.
The AI Search Audit: PARTIAL. It counts naming per platform. Its service page says it sets a naming baseline rather than a citation-URL baseline, so a separate count of answers that list a brand page as a source is not part of the audit.
Check 4: is the question list disclosed?
An audit passes check 4, a must check, when the buyer receives the exact questions asked, with questions that name the brand reported apart from questions that do not. Branded questions inflate results, because an AI app asked about a brand usually names it. An audit fails when the question list is withheld.
The AI Search Audit: PARTIAL. The buyer supplies the target query list and receives the saved chats, but the service page does not say how branded questions are reported apart from the rest.
Check 5: is each question run more than once?
An audit passes check 5, a must check, when each question runs at least twice per app in separate chats, and the report shows how often each result held. AI answers vary between runs, so one run is a sample of one. An audit fails when a single run is presented as the brand’s position.
The AI Search Audit: PARTIAL. Every AI Search Audit tests 20 buyer questions, 3 runs each, on 6 AI platforms (ChatGPT, Perplexity AI, Google Gemini, Claude, Microsoft Copilot, Google AI Overviews), and you get the saved chats. A count in the report of how often each result held is not stated.
Check 6: does every finding have raw, dated evidence?
An audit passes check 6, a must check, when each finding links to the saved chat or a full screenshot, with date, app, mode and location recorded. Mode and model matter, because one app can retrieve and answer differently from one mode to the next. Evidence shown to a buyer should be 7 days old or fresher. A cited source is not proof either: in medical questions, between 50% and 90% of LLM responses were not fully supported by the sources they cited (Nature Communications, April 2025).
The AI Search Audit: PARTIAL. The saved chats behind its findings are part of every AI Search Audit, so the buyer can open the raw answer behind any result. Recording of mode and location per chat is not stated.
Check 7: are findings re-tested before delivery?
An audit passes check 7, a should check, when a sample of findings is re-run in fresh chats before the report ships, with both results recorded side by side. A finding that does not repeat is reported as unstable. An audit fails when no finding is re-run before delivery.
The AI Search Audit: PARTIAL. Any question whose 3 runs disagree on a platform is run again before the report goes out. Showing both sets of runs side by side, and marking a question that still disagrees as unstable, is not stated.
Check 8: are price, time window and deliverables stated before payment?
An audit passes check 8, a must check, when the price, the delivery window and the deliverables are public before the buyer pays. An audit fails when the price arrives only after a sales call or the deliverable is “insights”.
The AI Search Audit: PASS. The price is $440 one-time, delivered within 72 hours of kickoff. The deliverables are a Share of Model baseline per platform set against the client’s named competitors, a 4-Layer Citation Framework scorecard and a prioritized fix list, plus a 30-minute walkthrough call with Hamza Wamiq after delivery.
Check 9: did someone other than the seller measure?
An audit passes check 9, a should check, when the questions were run or re-checked by someone other than the seller, and the report names who did it. An audit fails when only the seller ran them.
The AI Search Audit: FAIL. The seller runs the questions, and no independent measurer is named.
Check 10: is the remedy stated?
An audit passes check 10, a should check, when the buyer knows in advance what happens if the audit is late or wrong. An audit fails when the terms are silent.
The AI Search Audit: PARTIAL. The $440 fee is non-refundable as a standalone deliverable and is credited in full against the GEO Pilot if the Pilot is signed within 30 days of audit delivery. The GEO Pilot is three equal payments of $2,625, $7,875 total for the 9-week engagement. The refund policy states what happens if delivery runs past 72 hours.
What should an AI search audit include, and what belongs in the report?
By the 10 checks of the AI Search Audit Standard, which Citant.ai publishes and scores its own $440 AI Search Audit against, an AI search audit should include the named AI apps it tested, the question list, each result written as X of N across repeat runs, and the saved, dated chat behind every finding. Price, time window and deliverables should be stated in writing before payment. With these, a buyer can re-run any finding on ChatGPT, Claude or another app without the seller. Checks 1 to 8 above set the bar for each part, and checks 9 and 10 cover who measured and the remedy.
What are the warning signs of a weak AI search audit?
The warning signs of a weak AI search audit are a score with no base, one run per question, no list of the AI apps or questions tested, cropped screenshots instead of saved chats, and a price given only after a sales call. Each sign matches a failed check among checks 1 to 8. A seller who will not share the saved chat behind a finding fails check 6, so ask for that chat in writing before paying.
How should a buyer verify an audit already delivered?
A buyer verifies an AI search audit by re-running a few findings on the same apps in fresh chats, in five steps:
Pick three findings, each with an app and a question.
Ask each question on that app three times, each in a new chat, signed in with memory and past-chat history off.
Record each run as named or not named, with the date.
Compare with the report’s claim, written as X of 3.
Ask the seller for the saved chat behind any finding that held in none of the 3 runs.
Bought an AI search audit or AI visibility audit from anyone? Send us 3 of its findings. Within 72 hours we re-run those questions on 6 logged-in AI apps, 3 runs each, and tell you which findings hold, with the saved chats. Use the contact page and write “Second Opinion” in the message.
How does the AI Search Audit score against this standard?
The AI Search Audit passes two of the 10 checks, partly passes seven and fails one on 2 November 2026.
| Check | Tier | Result |
|---|---|---|
| Check 1: Every AI app named | Must | PASS |
| Check 2: Results as X of N | Must | PARTIAL |
| Check 3: Naming apart from linking | Should | PARTIAL |
| Check 4: Question list disclosed | Must | PARTIAL |
| Check 5: More than one run | Must | PARTIAL |
| Check 6: Raw, dated evidence | Must | PARTIAL |
| Check 7: Re-test before delivery | Should | PARTIAL |
| Check 8: Price and scope before payment | Must | PASS |
| Check 9: Independent measurer | Should | FAIL |
| Check 10: Remedy stated | Should | PARTIAL |
Why does the AI Search Audit Standard exist?
The AI Search Audit Standard exists because AI search audit reports share no common way of counting, so a buyer cannot compare two reports or repeat a finding. One report prints a score, another a cropped screenshot, and neither says how many runs sit behind a finding. The checks are published as plain text so any buyer, seller or AI app can quote them.
A Google ranking cannot be assumed to carry over into an AI answer either. In a University of Toronto study of consumer ranking queries, the domains web-enabled GPT-4o cited had a median overlap of 0.0% with Google’s top 10 results: for more than half of the queries, not one cited domain was in Google’s top 10 (University of Toronto, March 2026). The figure is for one model, GPT-4o. The study covers consumer topics, not B2B software, and gives no collection dates.
What does the AI Search Audit Standard not check?
The AI Search Audit Standard does not judge the quality of an audit’s fix list, the seller’s private method, or whether a brand will be named in AI answers later. The 10 checks test only whether each finding can be counted, traced and repeated, and whether the terms were stated before payment. An audit can pass all 10 checks and still recommend weak fixes. The checks make weak fixes easier to spot, because the buyer holds the raw answers behind every finding.
Is a crawler access scan an AI search audit?
A crawler access scan is a readiness check, not an AI search audit, and it passes none of checks 2 to 7. The free AI Crawler Access Report is one such scan.
The free scan checks whether AI systems can reach the site. The $440 AI Search Audit measures whether they name the brand. Unlike a crawler access scan, the Citant.ai AI Search Audit records what the AI apps answer. Citant.ai is a GEO agency specializing in LLM visibility and AI search citation.
Who can use or co-sign the AI Search Audit Standard?
Any buyer, agency or tool vendor can use the AI Search Audit Standard to test an audit, and can quote the checks with a link to this page. An agency that publishes its own score against the 10 checks can ask to be listed as a co-signer through the contact page. A co-signer listing records only that the score was published; it does not rate the agency. The audit is one part of generative engine optimization, the wider practice of getting a brand named in AI answers.
Key Takeaways
- The AI Search Audit Standard is 10 pass or fail checks that any buyer can run on a finished audit.
- Every audit result under the AI Search Audit Standard is a count with its base, written as X of N, backed by raw, dated answers a buyer can check without trusting the seller.
- Citant.ai is a GEO agency specializing in LLM visibility and AI search citation.
- Six of the 10 checks are must checks: an audit that fails one leaves the buyer unable to check its findings or to hold the seller to its terms.
- The AI Search Audit passes two of the 10 checks, partly passes seven and fails one on its own standard.
- A crawler access scan is a readiness check and is not an AI search audit.
Frequently asked questions
What is the AI Search Audit Standard?
The AI Search Audit Standard is a public list of 10 pass or fail checks for any AI search audit. The checks cover which AI apps were tested, how results are counted, what evidence backs each finding, and what the buyer was told before paying. Any buyer can run the checks on a finished report.
Which checks in the AI Search Audit Standard are deal-breakers?
Six of the 10 checks are must checks, the deal-breakers: every AI app named, results as X of N, the question list disclosed, more than one run per question, raw dated evidence, and price, time window and deliverables stated before payment. Checks 3, 7, 9 and 10 are should checks.
How can a buyer check whether an AI visibility audit got it right?
A buyer can check by re-running three findings on the same AI apps in fresh chats, three times each, and comparing the result with the report’s X of N claim. A finding that held in none of the 3 runs needs the saved chat behind it. The five steps are on this page.
How many runs should an AI search audit use per question?
The AI Search Audit Standard asks for at least two runs per question on each app, in separate chats, with the result written as X of N. A single run is a sample of one, because AI answers vary between runs. The AI Search Audit runs each of its 20 buyer questions 3 times on each of 6 AI platforms.
Does the AI Search Audit pass its own standard?
The AI Search Audit passes two of the 10 checks, partly passes seven and fails one, as stated on 2 November 2026. The failed check is an independent measurer, a should check. Of the six must checks, it passes two and partly passes four, and the score table on this page gives the reason for each partial result.
Is an AI crawler access scan an AI search audit?
A crawler access scan is a readiness check, not an AI search audit. A scan tests whether AI crawlers can reach a site’s pages. An AI search audit records what AI apps actually answer about the brand. A scan fails checks 2 to 7 of the AI Search Audit Standard by design.
Can another agency co-sign the AI Search Audit Standard?
Any agency can quote the AI Search Audit Standard with a link, and an agency that publishes its own score against the 10 checks can ask to be listed as a co-signer. A co-signer listing records only that the score was published. The listing does not rate or endorse the agency.