The terms behind the evidence.

The Absence Ladder groups measured visibility into published levels. Its result depends on the questions, answer base and thresholds stated with that measurement.

Answer engine optimization describes work on the information an answer engine can find and use. The name of the service does not replace evidence of whether a company was named or cited.

AI search visibility describes how a company appears in answers to a stated set of buyer questions. We report named answers, source citations and share of voice separately.

An answer engine turns a question into a written response rather than returning only a list of links. Its response and cited sources are separate pieces of evidence in our measurement.

A baseline is the dated first measurement used for an agreed comparison. Its questions, engine coverage and valid answer base define what a later measurement can be compared with.

A buyer question is a recorded question representing a software evaluation need. Its wording stays attached to the answer so another reader can inspect the scope of the observation.

ChatGPT is one of the five engine products named in our measurement. The report states the collection setup and does not claim to reproduce every buyer's individual screen.

A citation is a source link supplied with an observed answer. A company counts as cited as a source when a listed source belongs to its own domain or a confirmed alias.

Cited as a source means an answer includes a source on the company's own domain or a confirmed alias. An uncited mention is not inferred to be a source citation.

Claude is one of the five engine products named in our measurement. Its captured answers, valid base and exclusions remain identifiable in the evidence.

A confidence interval shows the statistical uncertainty around an estimated rate under the stated method. It does not remove limits caused by the question set, dates or measurement instrument.

Crawler access describes whether a request can reach a page under the observed access conditions. Our server's result does not prove how a provider's own verified request will be handled.

The denominator is the base against which a count or rate is reported. A named-answer rate and share of voice use different bases and must keep those labels attached.

Engine divergence is a difference in observed answers or vendor selections between engines. Our published study reports its own engine coverage, dates and answer base.

An entity is the company or other distinct subject being identified in an answer or source. Matching rules help separate that subject from similarly named organizations.

An excluded answer is an observation removed under a stated rule, such as a failed or empty response. We record the reason and do not count it as a measured absence.

A fixed panel is the agreed set of question and engine cells retained for a comparison. Extra adaptive answers do not enlarge that panel or silently change its denominator.

Generative engine optimization is work intended to improve how a company appears in AI generated answers. We assess it through recorded answers and sources, with the instrument and date stated.

Google AI Mode is a separate named search surface in our measurement. An observation from it is not substituted for an AI Overviews observation.

Google AI Overviews is one of the named search surfaces in our measurement. We report its captured answers and exclusions separately from Google AI Mode.

Indexing means a search service has included a resource in its searchable collection. A reachable page or an allowed robots rule does not by itself establish that inclusion.

An llms.txt file is a text directory of selected website resources with descriptions. It provides a readable map and does not guarantee that an engine will use or cite those resources.

An MCP connector gives a compatible assistant a structured way to request information from a service. Our public connector exposes published evidence with the scope stated on its connection page.

Named means the company appears in the answer text under the published matching rules. An answer counts once for that company even when the text repeats its name.

Perplexity is one of the five engine products named in our measurement. We report its answer observations and sources within the stated collection and publication limits.

A receipt is the evidence record behind an observation or reported figure. It connects the result to the captured answer and its available source references.

Recommendation position records where a company appears in a discernibly ordered or grouped shortlist. When the answer has no discernible order, we report that limit instead of inventing a position.

Record accuracy concerns classified statements about a company checked against dated public evidence. Unreviewed or unclassified statements do not imply either correctness or a zero error count.

Remeasurement repeats the agreed measurement after an interval or a change. We keep the comparison's scope and exclusions visible rather than treating unrelated samples as the same panel.

Robots.txt is a public file containing path access rules for named crawlers. Our access checks report those rules separately from whether a page is indexed or cited.

A run is one scheduled attempt to obtain an answer for a question on an engine. Repeated runs preserve differences instead of treating one response as a stable result.

Run-to-run noise is the variation between answers produced for the same question under the same stated setup. We retain the repetitions so a changed answer is not automatically presented as an improvement.

Share of voice is the company's named-answer count divided by the recorded brand mentions across the same valid answers. Each recorded brand counts once per answer, so this is a share of mentions rather than a percentage of answers.

A source map lists the published pages cited in a defined set of captured answers. Counts belong to that sample and do not claim to describe every answer an engine could produce.

Source presence records whether a company's own website appears among the sources for an answer. It does not mean that the answer also names or recommends the company.

A stale statement describes information that no longer matches the dated public evidence. It is distinguished from a statement whose truth cannot be checked.

Structured data labels facts in a machine-readable format alongside a page. Its presence can make those labels inspectable but does not establish that an answer engine will cite the page.

An unverified statement does not yet have enough recorded evidence or review to support a classification. We do not automatically turn that uncertainty into a wrong-statement count.

A valid answer is a captured response that meets the measurement's published inclusion rules. It contributes to the reported answer base for the relevant engine and question group.

A wrong statement conflicts with the evidence used for its recorded classification. Its supporting source and review status remain attached to the finding.