PRecious PulseBy PRecious Communications

What do the terms behind AI discovery actually mean?

Plain-English explanations and examples spanning AI answers, GEO, SXO, visibility, sources and measurement. Use them to make sense of the subject across your team.

Blue magnifying glass over an ivory reference card
AI visibilityWhether AI mentions
your brand—and gets
the facts right.
Use this page to decode terms, not to score your brand. Each definition has a quick example. For diagnosis and next steps, open the practical Q&A.

Answers and visibility

AI answer
A generated response to a user's question. It may summarize information, list options, give links or recommend an action.

Example: A buyer asks for SME banking options and receives a generated shortlist.

AI search
A search experience that uses AI to produce or organize an answer alongside or instead of a conventional list of links. Google AI Overviews, ChatGPT search and Microsoft Copilot have different interfaces and source conditions.

Example: ChatGPT search answers a question with linked sources instead of only a list of results.

AI visibility
How a brand, page or source appears in observed AI answers. A mention, first mention, citation and recommendation are different measures.

Example: A bank is mentioned in 6 of 10 saved answers in a defined test.

Answer accuracy
Whether an observed AI answer describes a brand or fact correctly when checked against dated, inspectable sources. Accuracy is separate from being mentioned or cited.

Example: A team checks an answer’s stated fee against the current fee page and records the answer date.

AEO, answer engine optimization
A label for improving the chance that useful, accurate information can support answers. It is a practice label, not an official platform score or a guarantee of inclusion.

Example: A service page answers a customer question clearly and cites its own evidence.

SXO, search experience optimization
Improving the full search journey: helping people find a relevant answer, understand the evidence and take a useful next step. It combines discoverability with clarity and usability; it does not guarantee an AI mention.

Example: A service page answers a buyer question, shows current proof and makes the next action clear.

GEO, generative engine optimization
A practice label for improving how useful, verifiable content may contribute to AI-generated answers. It is not an official platform score or guarantee.

Example: A team makes a claim page easier to verify across its site and independent sources.

SEO, search engine optimization
Improving a site's discoverability and usefulness in search. Technical access, helpful content and clear site structure still matter for Google's AI features, according to Google Search Central.

Example: A crawlable service page earns visits from conventional search results.

Mention and first mention
A mention is a brand name in the answer. First mention records which brand appears earliest. It describes answer order; it is not proof of preference, endorsement or purchase intent.

Example: An answer names Brand A before Brand B.

Recommendation
An answer that presents a brand as a suitable option for a stated need. It is stronger than a bare mention, but still needs evidence.

Example: An answer suggests Brand A for a stated business need.

SEO, AEO and GEO overlap. SEO covers discoverable, useful pages; AEO and GEO are practice labels for making information clearer and easier to verify as a possible input to AI answers. None guarantees a mention or citation.

Sources and proof

Citation
A visible source link or reference displayed with an answer. It shows a source was presented, not that every nearby claim is supported.

Example: An answer links to a bank fee page.

Grounding
Using retrieved or supplied material to support an AI response. Grounding can help trace evidence, but the generated wording still needs checking against the source.

Example: A generated answer uses a retrieved policy page for context.

Retrieval
The step where a system finds material to use in an answer. A page can be indexed yet not retrieved for a given question. The precise retrieval process varies by product and is not fully visible to site owners.

Example: A system finds an annual report before composing an answer.

Owned evidence
Facts and explanations on channels a brand controls, such as its website or published reports. Their quality affects how easily a reader can check a claim.

Example: A product page states who is eligible and what it costs.

Earned evidence
Independent reporting, expert commentary or other third-party material that can corroborate a claim. Its value depends on relevance and credibility; it cannot be ordered into an AI answer.

Example: An independent article verifies a launch date.

Proof gap
The distance between what an answer or brand claims and the evidence a reader can inspect. An example is a “regional leader” claim with no dated, market-specific support.

Example: An answer says “regional leader” but provides no dated support.

Claim-to-source fit
Whether a linked source actually supports the statement beside it. A link alone is not enough.

Example: A fee page supports a fee claim, not a security claim.

Crawling and access

Crawler
Software that requests pages for discovery or other stated purposes. OpenAI distinguishes OAI-SearchBot for search from GPTBot for training; permissions can be set separately.

Example: OAI-SearchBot requests a public product page.

robots.txt
A site file that tells compliant crawlers which paths they may request. A robots rule can affect discovery. It is not a promise that an allowed page will be cited.

Example: A Disallow rule asks compliant bots to skip a private path.

noindex
A directive asking search engines not to index a page. Use it deliberately. If a crawler cannot access a page, it may also be unable to read the directive.

Example: A staging page asks search engines not to include it in results.

Canonical URL
The preferred URL for substantially duplicate pages. It helps consolidate page signals but is a hint, not a command that forces an AI answer to use the page.

Example: Two copies of a guide point to one preferred URL.

Structured data
Machine-readable labels for visible page facts. Use supported types accurately. Google says no special markup is needed for AI features; FAQ markup is not a shortcut to AI recommendations.

Example: Article markup matches the visible article title and date.

Measurement

Prompt set
A documented list of realistic customer questions used for repeated checks. Include stage of decision, market, language, engine and test date. Avoid substituting brand-name searches for discovery questions.

Example: A test uses ten buyer questions about fees, trust and business needs.

Share of voice in a tested sample
The proportion of observed answers in a defined prompt set where a brand appears. State the denominator and sampling rules. It is not a share of all AI answers or all customer exposure.

Example: A brand appears in 6 of 10 saved answers: 60% of that sample.

Citation count
The number of times a source appears as a citation in a measured platform or sample. Bing's AI Performance report has citation metrics, with limits on what they show about answer placement.

Example: A source URL appears as a citation three times in a measured set.

Platform impression
An appearance recorded under a platform’s reporting rules. It is not a saved AI answer, a citation, a visit or a measure of exposure across all AI systems. Check the platform’s definition and reporting limits before comparing it with another measure.

Example: A Search Console impression is reported for a page shown in a Google search result; it does not show whether ChatGPT cited the page.

Referral traffic
Visits arriving from a source such as ChatGPT search. It shows click-through, not every answer exposure or mention. Check tagging and attribution rules before comparing sources.

Example: A visitor clicks from ChatGPT search to a brand website.

Qualified lead
An enquiry that meets agreed sales criteria, such as relevant need, market and ability to act. It is a business outcome to track separately from an AI answer appearance or a website visit.

Example: A visitor from an AI referral requests a briefing and meets the team’s agreed qualification criteria.

Market and language variation
Differences in answers when geography, local evidence or language changes. One market result need not describe another.

Example: A category question names different providers in Singapore and Indonesia.

From definition to decision

A term names a signal; it does not tell you what to fix. Use the Q&A to diagnose a specific answer and choose a practical next step.

Read the Q&A and 30-day plan, try the free AI Visibility Readiness Check to assess your team's approach and next action, or ask PRecious about a measured review.

Primary references