The CiteMoment 397-check catalog, explained

How to read the numbers

The CiteMoment catalog is the list of everything the audit is able to examine. It is published openly because a method you cannot inspect is a method you cannot trust. This page explains where the number 397 comes from, what each part of it means and what it does not mean.

The catalog is recorded in the product’s own check registry. The version this page describes is engine397-solution-contract-1.0, dated October 5, 2026. The registry is the source, and the tables on these pages match it entry for entry.

The arithmetic behind 397

The total is simple once the three layers are separated.

1. Site readiness: 93 records. These are checks of your website that do not depend on any one AI product. They are numbered T01 through T93.
2. Shared channel records: 180 records. There are 30 shared measurement definitions. Each one is applied to all six AI surfaces as a separate record, so 30 × 6 = 180.
3. Channel-specific records: 124 records. These checks exist for one surface only, because they depend on that product’s crawler, data path or behavior.

Add them up: 93 + 180 + 124 = 397. The shared and specific records together account for the 304 surface records, and the site records bring the total to 397.

Why the surface totals differ

Every surface carries the same 30 shared records, but the number of specific records differs because the products expose different things. Microsoft Copilot has 23 specific records because Bing Webmaster Tools, IndexNow and the NOARCHIVE and NOCACHE directives each add questions. Google AI Mode has 18 because it shares much of its technical foundation with Google Search and mostly adds multi-turn follow-up behavior. The differences are a property of the products, not a ranking of which one matters more.

Why there are 247 unique definitions

A record is a check applied to a particular place. A definition is the check itself. The shared definitions show the difference: “Raw answer and source record” is one definition, but it is recorded separately for ChatGPT Search, Google AI Overviews, Google AI Mode, Gemini, Copilot and Yandex, because the raw answer from each is different evidence.

Counting definitions instead of records gives 93 site definitions, 30 shared definitions and 124 surface-specific definitions, a total of 247. Both figures are legitimate, and CiteMoment is explicit about which one it is quoting. Quoting 397 as “397 checks” without the explanation would suggest more independent tests than exist, which is exactly the kind of inflation this catalog is designed to avoid.

What 397 is not

The catalog is an audit contract. It defines what the audit may examine and how each finding must be reported. In any given report some records are completed, some are partial, some cannot be measured and some do not apply. The report states the count of each, and it never lets an unverifiable item count as a pass.

The reasons a record may not be completed are ordinary. A provider may not disclose a condition such as the exact personalization state of a chat. A data path such as server logs may not have been shared. A check about product pages does not apply to a site that sells nothing. Reporting these cases separately is a feature: it shows you where knowledge ends.

How record IDs work

Each record has an identifier that tells you where it belongs and whether it is shared.

• T01–T93 are site-readiness records. Each one is also its own definition.
• CHATGPT-NN, AIO-NN, AIMODE-NN, GEMINI-NN, COPILOT-NN and YANDEX-NN are surface records. Numbers 01–16 are the shared COMMON-01 to COMMON-16 definitions applied to that surface, and numbers from 17 up are specific to the surface. The remaining fourteen shared definitions carry an M in the ID, for example CHATGPT-M17 to CHATGPT-M30, and repeat on every surface in the same way.
• COMMON-NN and COMMON-MNN are shared measurement definitions. They appear on this page in full.

Because every record has a stable ID, a finding in a report can refer to T10 or COPILOT-24 and you can look up exactly what that check examines. IDs also let a re-test confirm that the same check is being compared across periods.

How reports present the catalog

Because the catalog is a contract rather than a scorecard, a report presents it in four groups.

The four record statuses

Status	Meaning	Example
Completed	Evidence was obtained and the result is stated with scope, date and source.	T10: the robots rule that allows OAI-SearchBot is quoted.
Partial	Evidence exists but is limited by sample, period, market or access.	COPILOT-35: grounding queries are a limited sample.
Unmeasurable	The data is not disclosed or the data path is not available.	CHATGPT-24: the personalization state of a session is unknown.
Not applicable	The check does not fit the sector or page type; the reason is recorded.	T83: total cost and taxes on a page with no sales.

A report may show a score for the completed items, but that score always comes with its coverage. A score of 90 percent on 40 of 397 records is a very different statement from the same score on 300 records, and the report makes that visible.

The 30 shared measurement rules

The shared rules are the discipline that keeps measurements honest. The first sixteen, COMMON-01 to COMMON-16, define what must be recorded about every observation. The next fourteen, COMMON-M17 to COMMON-M30, define the methodology and reporting controls. Each rule is repeated on every surface because the evidence it governs differs by surface.

Read together, they say the same few things: record the conditions, keep the raw answer, do not mix paths, separate mention from recommendation, show denominators, show uncertainty, compare like with like, and re-test after fixing. The full list follows.

What the shared rules prevent

It helps to see the mistakes the rules are written to prevent.

• Mixing surfaces. Treating a developer interface result as if it were the consumer product, or the reverse, produces conclusions about a product nobody measured.
• Mistaking a mention for a recommendation. A brand named in a list of options is not a brand recommended. COMMON-08 keeps them apart.
• Counting listed sources as citations. A source in a panel that no sentence relies on is a different thing from a cited source. COMMON-09 and COMMON-11 separate the roles.
• Confusing namesakes. COMMON-07 and COMMON-M19 stop a similarly named business from being counted as yours.
• Reporting precision that does not exist. COMMON-M23 and COMMON-M24 force denominators and uncertainty into view.
• Declaring victory without a re-test. COMMON-M30 requires a repeat measurement and a regression watch after a fix.

How the site layer and the surface layers work together

The 93 site checks answer whether AI systems can reach and understand your pages in the first place. The surface records then ask what each system actually did with them. The two layers need each other.

If a surface measurement shows that your site is never cited, the site layer is the first place to look: is the crawler allowed, is the page indexed, is the content visible without scripts, is the entity clear? If the site layer is clean and the surface measurement still shows no citation, the cause is more likely relevance, authority, competition or simply the way that surface chooses sources. Without the site layer you would not know which situation you are in.

Likewise, a perfect site layer does not prove visibility, because an AI answer is generated on demand and cannot be predicted from site hygiene alone. That is why the catalog includes dozens of records whose only job is to separate real visits from citations, requested search from performed search, and listed sources from cited ones.

See the complete lists on the site readiness checks page, the ChatGPT, AI Overviews and AI Mode page and the Gemini, Copilot and Yandex page.

Using the catalog without buying an audit

You can use this catalog as a self-review checklist even if you never commission an audit. A sensible order is the following.

1. Start with crawl access. Check T09 to T14: robots rules, search crawler permissions and whether a firewall blocks real bots.
2. Confirm indexing signals. Check T15 to T25: status codes, redirects, canonicals, noindex, snippet limits and the sitemap.
3. Read your own pages as a stranger. Check T26 to T36: does each page answer its question directly and clearly?
4. Review facts and identity. Check T46 to T57: language, names, addresses, phone numbers and structured data.
5. Look for contradictions. Check T82 to T88: stock, price, language versions and stale copies.
6. Only then look at AI answers. Ask real customer questions on each surface and save the answer, the sources and the date.

A self-review will find obvious problems. What it cannot easily do is verify bot identity against official records, combine access evidence with server logs, repeat measurements under controlled conditions or classify hundreds of answers consistently. That is where a structured audit earns its keep.

A worked example: following one question through the catalog

Imagine a dental clinic that asks, “Why does an AI assistant keep giving the wrong opening hours for us?” Here is how the catalog turns that complaint into a structured investigation.

1. Capture the answer properly. The shared rules (COMMON-06, COMMON-M17 and COMMON-M27) require the full answer, its sources, the exact question, the conversation version and the timestamp to be stored, so the wrong hours can be inspected later and not only remembered.
2. Identify the entity. COMMON-07 and COMMON-M19 confirm that the answer is really about this clinic and not a clinic with a similar name in another city. T49 and T48 check how the site separates namesakes.
3. Find the origin of the claim. COMMON-11 matches the sentence with hours in it to its cited source. If the source is a directory listing, T84 compares opening hours across the profile, the owner’s records and the website, and T91 records the wrong external record with the person who can correct it.
4. Check the site’s own contribution. T55 and T52 compare the hours and phone number in the visible text with those in structured data, while T88 searches for a stale version of the page left in cache.
5. Fix and re-test. COMMON-M29 matches the finding to a fitting fix, and COMMON-M30 requires the same question to be run again under the same conditions, with a regression watch afterward.

Notice what this example does not need: a single magic score. It needs a trail of evidence, an owner for each correction and a repeatable way to confirm the result. That is the real purpose of a catalog with a stable ID for every record.

The same approach works for other complaints. If an answer quotes an old price, GEMINI-19 or COPILOT-21 trace the source passage on those surfaces. If a brand is never cited, the site layer is examined first, beginning with T09 to T14. If citations exist but visits do not, CHATGPT-26 and CHATGPT-35 separate visits from citations using your own analytics.

Versioning and updates

AI products change quickly. Crawler names appear, data panels are added and documented behavior shifts. The catalog is versioned so that a report can state which contract it used, and so a later report can show whether differences come from your site or from a changed catalog. Where a check depends on current provider behavior, the report refers to a dated source rather than to memory.

When the catalog changes, the changes are visible: records are added, retired or reworded with a new contract version, and old reports keep the version they were produced under. Nothing is quietly rewritten.

Frequently asked questions about the catalog

Why does the catalog have 397 records but only 247 unique definitions?

Thirty shared definitions are applied to each of six AI surfaces as separate records. Thirty applied six times gives 180 records. Adding the 93 site definitions and 124 surface-specific definitions gives 247 unique definitions and 397 records.

Are the 397 records official ranking factors?

No. The catalog is an audit contract. 397 is not an official count of ranking factors, and no search provider has published such a list.

Does every report complete all 397 records?

No. Each report separates completed, partial, unmeasurable and not-applicable records. A record that cannot be verified with real data is not counted as passed.

What do the record IDs like T10 and COPILOT-24 mean?

T-numbers are site-readiness records. IDs that begin with a surface name, such as COPILOT-24, are records for that surface. Numbers 01 to 16 and IDs with an M, such as COPILOT-M17, are shared rules, and plain numbers from 17 up are specific to the surface.

Can I use the catalog as a checklist myself?

Yes. The site-readiness checks and shared rules work as a self-review list. A structured audit adds verification against logs and official records, repeated measurements and consistent classification.

How often does the catalog change?

It is versioned. The version described here is engine397-solution-contract-1.0, dated October 5, 2026. Reports state the contract version they used, and changes are made visible rather than rewritten silently.