An AI search visibility baseline is a documented view of when, where and why a brand appears in AI-generated answers. It gives a team something more useful than a one-off screenshot: a repeatable starting point for deciding what to improve and measuring whether the work has changed real discovery outcomes.
What an AI visibility baseline should answer
A useful baseline does not try to reduce every observation to one headline score. It should answer a small set of practical questions:
- Which commercially important topics and questions are being monitored?
- Does the brand appear directly, appear through a cited page, or remain absent?
- Which competitors or alternative sources are consistently mentioned?
- What evidence appears to support the answer?
- Which weaknesses can the organisation realistically address?
These questions keep the exercise connected to customer journeys rather than treating AI visibility as an isolated vanity metric.
1. Define topics before writing prompts
Begin with the decisions your audience is trying to make. A software company might group demand around problem education, category comparison, implementation, security, pricing and vendor selection. A professional-services firm may organise topics around symptoms, legal or commercial implications, service options, proof of expertise and location.
For each topic, write prompts representing different stages of intent. Include exploratory questions, comparison questions, risk questions and action-oriented requests. Avoid producing dozens of minor wording variations before the core decision journeys are covered.
A practical starting set
Use five to ten priority topics, with three to five prompts for each topic. This is usually enough to reveal meaningful patterns while remaining manageable for human review.
2. Audit the sources AI systems can understand
Before checking answers, examine the evidence available on the website and across credible third-party sources. AI systems cannot reliably represent expertise that is vague, fragmented or unsupported.
Review whether each priority topic has a clear destination page, concise definitions, evidence of experience, identifiable authors or reviewers, relevant structured data, internally linked supporting material and consistent organisation details. Then review external mentions, citations, profiles, reviews and specialist references that help corroborate important claims.
3. Observe the answer set consistently
Run the agreed prompts using a documented method. Record the platform, date, prompt, response summary, brand mention, linked citation, competitor appearances and any uncertainty. Because generative answers can vary, a single response should be treated as an observation rather than a permanent ranking.
Repeat high-priority prompts and keep the conditions as consistent as reasonably possible. The goal is not to manufacture precision. It is to identify recurring patterns across a controlled sample.
4. Separate mentions from citations
A brand mention and a source citation are related but different outcomes. A system may name a company without linking to it, or cite one of its pages without making the brand prominent in the answer. Track both.
| Signal | What it indicates | What to review next |
|---|---|---|
| Brand mentioned | The organisation is recognised as relevant to the prompt. | Accuracy, context and competitor positioning. |
| Website cited | A page is being used as supporting evidence. | Page quality, extractability and conversion journey. |
| Competitor mentioned | Another entity is more strongly associated with the topic. | Content depth, authority signals and differentiated proof. |
| No relevant appearance | The available evidence may be insufficient or poorly aligned. | Topic coverage, entity clarity and external corroboration. |
5. Connect observations to actions
The baseline becomes valuable when every important gap has an owner and a next action. Some gaps will require clearer service or product pages. Others may need expert-led supporting content, stronger internal linking, corrected structured data, better author information or credible third-party validation.
Prioritise work using commercial importance, frequency of the observed gap, effort and confidence. Keep observed data separate from interpretation, and label estimates clearly.
How often should the baseline be refreshed?
High-priority prompts can be reviewed monthly, while a broader topic set may be revisited quarterly. Refresh the baseline after major site changes, positioning updates, product launches or significant changes in the search platforms being monitored.
Historical comparison matters more than constant checking. A stable review rhythm makes it easier to distinguish a durable change from normal response variation.
Common mistakes to avoid
- Tracking only the company name instead of unbranded customer questions.
- Using hundreds of prompts without a clear link to audience intent.
- Reporting an opaque score without preserving the underlying observations.
- Treating one generated answer as a fixed search result.
- Ignoring competitor mentions and the sources supporting them.
- Publishing more content before correcting weak entity and evidence signals.
A focused 30-day starting plan
- Week one: agree priority topics, decision stages and a controlled prompt set.
- Week two: capture observations and audit the owned pages connected to each topic.
- Week three: compare recurring competitors, citations and missing evidence.
- Week four: assign the highest-confidence technical, content and authority actions.
Build a baseline people can trust
AI search visibility is still an evolving measurement area. The most credible approach is transparent about what was observed, how it was collected and where judgement was applied. A smaller, well-documented baseline is more useful than a large dashboard that cannot explain its numbers.
Visibility North combines AI visibility observations with search performance, technical evidence, content intelligence and authority signals so teams can move from interesting outputs to accountable action.
