AI search glossary

Answer engine optimisation. Work on how information appears in direct answers, including AI answers.

An AI answer shown within Google Search. Record whether it appeared and which sources it linked.

A label for AI search optimisation or using AI tools for SEO. Ask which meaning is intended.

How your business appears in the AI answers you check. Those answers are a sample of AI search.

A system that responds to questions with answers. It may also show source links or search results.

An answer collected through a programming interface. It can differ from what the consumer app shows.

The method that links a visit or outcome to a recorded source. Say what it can't see.

The starting record you keep before a change. It needs a date and the same measurement rules.

A question that names your business. Keep these separate from questions that do not name it.

A detail your business can support with a current record and a named owner.

Readable answers divided by requested attempts. Show missing captures separately.

One change producing another. A before and after sequence alone can't show it.

A source link attached to an answer. Check which statement the linked page actually supports.

A rate calculated from a stated citation rule and a stated total. A different rule gives a different rate.

The app or website a person uses to ask an assistant. Record its context separately from an API.

A comparison kept apart from the change being tested. Its design affects what you can conclude.

Software that requests website pages. A crawler visit isn't a mention, a citation or a customer visit.

The total you divide by when calculating a rate. Name the actual population it counts.

Generative engine optimisation. Work on content visibility in generated answers.

Connecting an answer to supplied or retrieved information. Check that the cited source supports the statement beside it.

A search system adding a page to its index. Being eligible and being chosen are different things.

The kind of AI model behind assistants such as ChatGPT, Gemini and Claude. It generates text from learned patterns and the context supplied. Web access depends on the tools it is given.

An answer names the business. It may name it without linking to the business website.

Answers naming a business divided by readable answers, under the stated counting rule.

An attempted observation that produced no readable answer. It doesn't show your business was left out.

A question that does not name your business. It can help review answers about a customer need.

The difference between two percentages. Twenty to thirty percent is a change of ten percentage points.

The question or instruction sent to an assistant. Save the exact wording and its context.

Related searches an AI feature may run to build an answer. Your record may not show them.

The group of questions included in a review. Keep its version when adding or removing questions.

A recorded website visit from a source. It isn't a crawler request or an answer mention.

Turning page code into the content a reader sees. Check which content the tested client can read.

A file that gives crawlers instructions. It doesn't show that a page was read or used in an answer.

The observations included in a review. It won't cover every customer or every possible answer.

Search engine optimisation. Work that helps a page get crawled, indexed and shown in search results such as Google and Bing.

The places a business says it serves. A base address and a service area are different facts.

A comparison under a stated counting rule and competitor set. Always show which businesses and answers were included.

A difference between the sources seen in two groups of answers. It doesn't explain why they differ.

Machine-readable labels describing page content. Check that they agree with the visible information.

The particular app or search experience where an answer was observed. Keep it with the provider name.

The period included in a measurement. Keep the dates and time zone with the count.

A prompt selected for repeated checks. Each tracked prompt is checked every day on the three standard assistants.

The evidence for a figure is missing. That's different from a measured zero.