Auditing & diagnosis

A complete audit answers four questions:

1. Visibility. Does the brand appear in answers, and how often compared with competitors?
2. Brand Knowledge. How confident are assistants when answering questions about the brand?
3. Sentiment. How do assistants describe the brand, and on which topics?
4. Sources. Which sites do assistants rely on, and does the brand appear there?

In CiteMind, onboarding a client URL runs this baseline for visibility, Brand Knowledge, sentiment and revenue leakage.

There is no single right number. A focused set is easier to review and explain. As a starting point, many agencies begin with a set that covers each question type and the two or three most relevant assistants, then expand once the baseline is clear.

Cost scales with volume. In CiteMind, one monitoring run is executed each day for every tracked question and selected assistant, and you pay for the runs you schedule. A quick way to estimate monthly volume:

Monthly runs = questions x assistants x days

Example: 20 questions x 3 assistants x 30 days = 1,800 monthly runs.

Configure each client independently, so a small client does not carry the cost of a large one.

• Keyword-style questions. Assistants respond to natural questions, so write them as buyers would.
• Only branded questions. They tell you little about discovery. Include category and comparison questions.
• Too many questions at once. A large set is hard to interpret. Start focused.
• Changing the set every month. You lose the trend. Change questions deliberately and note the date.
• Ignoring sources. Without them you cannot explain why results look the way they do.

AI Visibility. How present the client is in AI answers across the tracked questions, including its share of voice compared with competitors. Use it to answer: are we showing up?

Brand Knowledge. How confident assistants are when answering user questions about the client, segmented by category. Low confidence in a category points to a weak area to improve. Use it to answer: do assistants know us well enough to answer correctly?

AI Sentiment. How assistants describe the client, shown by category such as Product Quality, Customer Support, Trust, Performance, Delivery, Value, Reliability and Pricing. Use it to answer: what do assistants say about us?

Together they separate three different problems: not being present, not being known and being described poorly.

Overall scores hide the useful detail. Look at categories.

• A client may have strong Product Quality sentiment but weak Pricing sentiment. That tells you to work on pricing information and perception.
• Brand Knowledge may be strong in one product line and weak in another. That tells you where to improve content and sources.
• Visibility may be high on branded questions and low on category questions. That tells you the brand is known but not discovered.

Always ask: which category is weakest, and which one matters most to revenue?

Assistants behave differently, so scores can differ between them. A client may be visible in one assistant and nearly absent in another.

• Compare assistants to find where the gap is largest.
• Prioritize the assistants your client's audience uses most.
• Do not average away differences. A strong score in one assistant can hide a weak one that matters.

Answers vary, so a single score on a single day is a weak signal.

• Compare the same question set across weeks.
• Look for sustained direction rather than daily swings.
• Note events that can move results: model updates, site changes, new content, competitor launches.
• Set expectations early that small movements are normal.

Do not show every number. Lead with:

1. One headline. For example: "Assistants recommend two competitors more often than you on comparison questions."
2. The category gap that most affects revenue.
3. One sentiment issue with a clear cause.
4. The top three actions you recommend, ranked by impact and effort.

Keep the full data available for questions.

• Scores describe a defined question set. They are not a measure of every possible question.
• Assistants change, so comparisons are most reliable within the same period and setup.
• A score is a signal to investigate, not proof of cause. Check sources and answers to understand why.

A score on its own is hard to judge. A benchmark gives it meaning. Clients respond to competition, and it helps you:

• Show where the client is behind and where it leads.
• Choose which questions deserve the first actions.
• Learn what the leading competitors have that your client lacks.
• Frame the work as an opportunity to gain share, not as a failure.

Use three to five competitors, a mix of:

• Direct competitors the client already knows.
• Competitors assistants recommend that the client may not have listed. Add any that appear often in answers.
• An aspirational leader in the category, if useful for comparison.

Ask the client for their list, then add what the assistants show. Surprises make good discussion points.

• Share of voice. How often each brand appears across the tracked questions, shown as a share of all mentions.
• By question type. Category, comparison, problem and decision questions often tell different stories.
• By assistant. A competitor may dominate one assistant and not another.
• By category of sentiment. Compare how assistants describe pricing, support, trust and reliability for each brand.
• Position in the answer. Whether a brand appears early or late in the response.

The most useful part of a benchmark is the reason behind the gap. For each question where a competitor beats your client:

1. Note the sources the assistant cites.
2. Check whether the competitor appears on those sources.
3. Check whether your client appears there, and how.
4. Identify the type of source: a comparison page, a review site, a community thread, a video, a media article or the competitor's own site.

CiteMind's Reasoning Chain Transparency shows why an LLM cited specific sources, which helps you defend your recommendations.

Translate each major gap into a concrete opportunity.

• Missing from a comparison question -> create or improve a comparison page, and check third-party comparisons.
• Weak on a sentiment category -> clarify the information assistants use on that topic, such as pricing, support or reliability.
• Competitor present on a source your client is absent from -> plan presence on that source.
• Strong competitor in one assistant only -> check what that assistant relies on and whether crawlers can access your client's pages.

Rank the opportunities by impact and effort, and pick a short list for the first month.