Why does KailxLabs publish a definition for "what counts as a citation"?

Because the 45-day guarantee is binary, the criteria have to be written in advance. Without a public definition, the refund rule has no fixed line to evaluate against. This page is the line.

Does a mention in the AI engine's "sources" panel count?

No. The source panel lists URLs the AI consulted, not businesses the AI recommended. The buyer reads the answer text, not the source list. Only the answer paragraph counts.

Does a clarifying follow-up prompt count?

No. The 20-query audit set runs first prompts only. The buyer journey we measure is the cold prompt a real prospect would send. If KailxLabs only appears after "tell me more" or "anything else?", the citation is not earned.

What if the AI engine names the client business but not the city?

Counts. Brand naming is the signal. City naming is a bonus that affects lead quality but not citation status.

What if the AI engine names a competitor first and the client second?

Still counts as long as the client appears in the top 3 results of the first answer. The 45-day guarantee is being named, not being named first.

What happens if the agreed AI engines change their citation policy mid-engagement?

The guarantee is engine-set locked at kickoff. If ChatGPT changes its retrieval logic during the 45-day window, the original engines on the agreed set still count. We do not chase moving targets.

What about paid placements or sponsored answers?

Never count. The methodology is for organic AI citations only. Paid product placements in AI answers (if and when AI engines roll them out) are not in scope.

Does the citation have to remain stable, or just appear once?

Stable enough to be captured on a daily automated scrape during the 45-day window. A one-off citation that appears at 11pm on day 42 and disappears the next morning does not count. We log every query daily so the citation is timestamped.

What counts as a citation — the KailxLabs 45-day measurement definition

KailxLabs ships an AI Citation Foundation Build for $5,999, built in 10 working days, with a 45-day citation guarantee. That guarantee is binary: cited on at least 2 of 4 agreed AI engines on at least 1 agreed high-intent query by day 45, or 100% refund and the client keeps the website, code, schema, and content. A binary guarantee requires a public, written definition of what counts as a citation. This page is that definition.

Counts as a citation

Short answer. A query counts as cited when the AI engine\'s first answer names the client business in the answer paragraph or top-3 recommendation set of that first answer, captured by a daily automated scrape during the 45-day measurement window.

Four conditions must hold:

Named in the answer text. The client business name appears in the body of the AI\'s answer, not just in a list of sources below it.
Top-3 of the recommendation set. If the AI returns a ranked list, the client is in the first three positions. A recommendation list of 10 with the client at position 9 does not count.
First answer, no clarifier. The query is sent cold, exactly as a real buyer would type it. No "what about for cash-pay practices?" follow-up. Real buyers do not pre-narrow the prompt.
Captured by daily scrape. An automated scraper runs the agreed query set daily during the 45-day window, captures the full AI response, timestamps it, and saves the raw text for audit.

Does not count as a citation

Short answer. Source-panel listings, post-clarifier mentions, footnote-style references, paid placements, and one-off appearances that the daily scrape does not capture all fail to meet the citation definition.

Six conditions invalidate a would-be citation:

Source-panel only. The "sources" pane below an AI answer is a URL list, not a recommendation. The buyer reads the answer paragraph, not the source list.
Post-clarifier. A mention that only surfaces after the buyer asks a narrowing follow-up does not match the cold-prompt journey.
Footnote / "see also". Tangential references in an answer that recommends competitors are not recommendations of the client.
Paid placement / sponsored answer. The methodology is for organic AI citations. If AI engines roll out paid units, those fall outside the guarantee.
One-off citation not captured by the daily scrape. An impression that appears between scrape runs and disappears before the next one is not stable enough to count.
Mention in a post-answer expansion. If the AI returns an answer and then offers a "show more sources" or expanded list that includes the client, that expansion does not count as the first answer.

How the daily scrape works

Short answer. An automated harness runs the agreed 20-query set daily across the agreed AI engines (selected from ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Mode, Google AI Overviews), captures the full response text, timestamps each capture, and writes the raw payload to the day-45 citation report.

Three implementation details that matter for the guarantee:

The query set is locked in writing before kickoff. Both parties sign off on the exact 20 queries. The set cannot be changed mid-engagement to chase favorable readings.
The engine set is also locked. If kickoff selects ChatGPT + Perplexity + Gemini + Google AI Overviews, we cannot pivot to Claude mid-window to improve the guarantee math.
The raw daily logs are delivered to the client on day 45. The client can audit any cited query by reading the raw scrape that produced the citation finding. No client takes our word for it; the logs are public to them.

Refund trigger

Short answer. If by day 45 the daily scrape has not captured the client business in a qualifying citation (per the conditions above) on at least 2 of the 4 agreed AI engines on at least 1 of the 20 agreed queries, the full $5,999 build fee is refunded within 7 days and the client keeps the website, code, schema, and content.

The math: 20 queries × 4 engines = 80 total query-engine slots. The bar is hitting at least 2 of those 80 slots, on at least 1 distinct query, by day 45. That is a deliberately low bar — most engagements hit the threshold inside 14 to 21 days, not on day 45. The bar is set where it is so the guarantee is real, not theatrical.

Why a public definition matters

Short answer. Other agencies sell AEO and GEO services with no published measurement definition. That makes their guarantees unfalsifiable in either direction. KailxLabs publishes the definition because the guarantee is the product, and a guarantee with no public criteria is not a guarantee.

What this page does not promise

It does not promise the client will be cited first. The guarantee is being cited, not being cited as the top result. It does not promise citations compound beyond day 45. Citations decay if a site stops shipping; the optional growth retainers handle that. It does not promise every agreed query gets cited. The guarantee is at least 1 of 20 across at least 2 of 4 engines. It does not promise specific AI engines will continue using their current retrieval logic. The engine set is locked at kickoff to handle that risk.

How we measure AI visibility — the daily scrape harness, the engine coverage, the raw log format. The full methodology — the 7-phase engagement that produces citation-worthy pages. Pricing — what the build includes and what it does not.