AI Search teams: don’t conflate visibility and citations.

The Jetpack team has attended a fair few AI Search conferences over the past six months. In among the genuinely useful sessions on structured data, entity grounding, and prompt design, we’ve noticed a recurring problem: speakers using “visibility” and “citations” as if they mean the same thing. They don’t. And the confusion has consequences for how brands report on AI Search performance, and what they invest in to improve it.

The conference observation

At one recent event, a well-regarded speaker showed a chart labelled “Brand Visibility” and then, minutes later, referenced the same numbers as “our citation rate.” The audience nodded along. Another session presented a case study where the KPI was “AI Search visibility,” but the tactics recommended — schema optimisation, entity refinement, deep-link content — were all clearly aimed at improving which pages get pulled as sources, not brand mention.

This is not nitpicking terminology. When a marketing lead reports “visibility is up 40% quarter on quarter” to their board, that number needs to correspond to something specific. If half the team means “brand mentioned in AI responses” and the other half means “our URLs cited as sources,” the reporting is measuring two different phenomena and calling them the same thing.

Defining the two terms

Visibility is whether your brand — as a named entity — appears anywhere in an AI-generated response. It’s the answer to “did ChatGPT mention Aviva when someone asked about pension consolidation?” Visibility measurement is comparatively coarse: the brand is either present or absent in the response text, at some position, in some framing.

Citations are which specific URLs the LLM pulled from to construct the response. It’s the answer to “which of Aviva’s pages appeared as source material?” Citation measurement is granular: it tracks page-level attribution, not brand-level presence.

How they diverge in practice

In traditional search these metrics ran in parallel — if your page ranked, it was visible; if it was visible, it drove traffic. In AI Search they decouple. Three patterns we’ve seen consistently:

Pattern one: mentioned but not cited

A financial services brand runs a well-known consumer campaign. When asked about pension providers, ChatGPT frequently names them. But when asked about specific pension technicalities, ChatGPT pulls source URLs from Which?, Money Saving Expert, and the FCA — not the brand’s own site. High visibility, low citations. The brand’s reputation is doing the visibility work; their content isn’t earning the citations.

Pattern two: cited but not visible

A specialist B2B SaaS company publishes deep technical documentation. LLMs pull chunks from their docs as source material for niche technical questions. But the brand is rarely named in the response — the LLM extracts the technical content without attributing it to a specific vendor. High citations on a narrow topic set, low overall brand visibility.

Pattern three: divergent by engine

The same brand can be highly visible in Perplexity but nearly invisible in Google’s AI Mode. Different LLMs weight brand mentions and source pulls differently. A brand cited by Perplexity might not appear in Gemini responses at all, even for identical prompts. Measuring at the aggregated “AI visibility” level obscures these engine-level patterns entirely.

Optimisation implications

The tactics that improve each metric are almost entirely separate. Improving visibility is a brand and entity problem: get named in Wikipedia and Wikidata, earn mentions in high-authority category coverage, build knowledge graph presence. Improving citations is a content and structure problem: publish deep, well-structured content on the specific queries you want to be cited for, mark up entities correctly with schema, keep pages current.

If a marketing team reports “visibility” as their KPI and then invests exclusively in content optimisation — the citation lever — they will see modest improvements at best, because they’re pulling the wrong lever for the reported metric. Similarly, an entity-optimisation programme won’t move citation numbers if your content itself isn’t earning source attribution.

The correct move is to report both, distinctly, and to identify which one your programme is genuinely optimising for. Both matter. Neither substitutes for the other.

Measuring both distinctly

GEO Jetpack measures visibility and citations as two separate metrics on every prompt, across every engine. On the visibility side we track brand mention rate, position, and framing within the response. On the citation side we extract every URL the LLM pulled from and score authority, source type, and whether the citation belongs to the tracked brand or a competitor.

This separation matters because it means the dashboards, reports, and ENQ™ recommendations tell you which metric is moving and why. If visibility is up but citations are flat, we surface the entity-optimisation work that would explain the gain. If citations are up but visibility is unchanged, we point at the specific content decisions driving the citation growth.

The takeaway

If you take one thing away from this piece, make it this: the next time someone at a conference (or in your team) says “AI visibility is up,” ask them which metric they mean. If they don’t have a clear answer, the reporting isn’t rigorous enough to base investment decisions on. And if they do have a clear answer, the follow-up question — “and which of the two is your team actually optimising?” — is the one that matters most.

The AI Search space is still forming its measurement conventions. It’s on practitioners to insist on precise vocabulary now, before imprecise reporting becomes normalised. Two words, two metrics, two different sets of tactics. Keep them apart.