A Simple Monthly Routine for AI Share of Voice

AI share of voice sounds like a precise metric, but answer-engine outputs are variable. The response can depend on wording, location, freshness, account context, retrieval availability, and the engine's current product behavior. A useful tracking routine therefore measures a repeatable set of observations rather than claiming to know every recommendation a customer could receive.

The purpose is to spot patterns: whether your business is understood, whether it appears for important questions, which sources are used, and where competitors or alternatives are described more clearly.

Define what you are measuring

Write a short definition before collecting data. For example, your monthly view might include whether the business was named in an unbranded answer, whether the relevant page or profile was cited, whether the description was accurate, and which alternatives appeared. Keep "named" separate from "cited." A business can be mentioned without a useful source, and a source can be cited without producing a recommendation.

Do not combine every service, location, and question into one number. Separate the dimensions that matter to the business: emergency service versus planned work, city center versus surrounding areas, or homeowners versus commercial customers. A small, meaningful set is easier to repeat and interpret.

Build a fixed prompt set

Choose a group of natural questions that customers actually ask. Include a broad category question, a local provider question, a problem-specific question, a comparison question, and a question about what to look for. Add follow-ups that test whether the system can explain a recommendation accurately.

Remove your business name from the discovery prompts. Include the location and relevant situation, but do not make the wording so specific that only one business can qualify. Save the exact spelling, punctuation, location context, language, date, and device or account conditions when those factors matter.

Keep a separate branded verification prompt. It tests whether public information about your business is coherent; it should not be counted as proof of discovery visibility.

Record the full response, not a screenshot alone

Create a spreadsheet with one row per prompt and date. Useful fields include engine or experience, prompt, location context, businesses named, source URLs, whether your business appeared, description accuracy, relevant page cited, and notes about uncertainty. Save the full response or a durable text copy when permitted by the product's terms. A screenshot can supplement the record, but it may not capture links or the entire answer.

Mark answers as accurate, partly accurate, inaccurate, or not enough information. Note specific errors: wrong service area, old hours, a service not offered, confusion with a similarly named company, or a broken source. This detail is more actionable than a simple yes or no.

Use a modest score only if it helps decisions

If your team wants a summary, create a transparent internal index. For each discovery prompt, you might mark appearance, accurate description, and relevant citation as separate yes-or-no fields, then report the counts alongside the raw table. Do not present that internal index as a universal market share or an official engine ranking. Avoid blending engines whose interfaces and source behavior are materially different.

A range or trend can be more honest than a single precise number. If the results are sparse, say so. If a business appears in one run and not another, record the variation instead of treating it as a permanent gain or loss.

Diagnose before changing content

When the business is missing, first ask what kind of problem the observation suggests. Is the service or location unclear? Is the page inaccessible or hard to discover? Are profiles inconsistent? Does a cited directory describe an old service? Does the business have useful evidence but no page that answers the question directly?

Make one or two related improvements, assign an owner, and document the date. Possible actions include rewriting a service opening, correcting a profile, clarifying a service area, adding a process explanation, or fixing an internal link. Avoid changing the entire site after one surprising response; you will not know what affected later observations.

A repeatable monthly calendar

During the first week, run the fixed prompts under the same practical conditions and record the raw answers. During the second, classify errors and inspect cited sources. During the third, make a small set of verified updates. During the fourth, review whether the changes improved clarity for a human reader and whether the next run provides new evidence.

Keep a change log with URL, exact edit, reason, approval, and publication date. At the next monthly review, compare the prompt wording and conditions before drawing a conclusion. If you changed the location, account, engine feature, or prompt, label the comparison as imperfect.

Share a short report with three sections: what was observed, what is uncertain, and what will be done next. Include examples of accurate and inaccurate descriptions so the team understands the goal. The report should lead to maintenance work, not panic or promises.

Common tracking mistakes

Do not run only branded searches, count every citation as an endorsement, or compare a fresh answer with an old one without recording context. Do not ask employees to repeatedly prompt an engine until a favorable result appears. Do not publish a score without explaining its prompts, date range, and limitations. Do not use a competitor's appearance as evidence that their business is universally preferred.

A careful routine cannot reveal a hidden, exact market share. It can show whether your public information is understandable, where engines retrieve it, and which customer questions remain unanswered. That is enough to prioritize responsible improvements.

Over time, the value is the history: a consistent prompt set, source notes, correction log, and honest interpretation. Treat AI visibility as an observable part of the customer journey, not a permanent position that can be guaranteed or purchased.