5 Mistakes to Avoid When Measuring Your AI Visibility

If you are starting to measure how ChatGPT, Perplexity, or Gemini talk about your brand, you are already ahead of most of your competitors. But there are 5 mistakes we see coming up constantly from brands just starting to track their AI visibility.

Each one is avoidable. And each one has a direct impact on the quality of insights you will get.

Mistake 1: Prompts That Do Not Represent Your Market

This is the most frequent and most costly mistake. If your prompts do not reflect the reality of your market, every analysis that follows will be skewed.

A good prompt set must cover the entire decision journey of your customers. From the prospect who does not yet know your category ("How do I choose the right home insurance in Switzerland?") to the one actively comparing ("Helvetia or Zurich, which one should I choose for an SME?").

The Two Most Common Errors

Too many branded prompts, not enough industry prompts. If you only track prompts that mention your brand, you will never see the competitive landscape. Industry prompts (those describing a need without naming anyone) are where the real battle takes place: when a prospect does not know you yet, the AI's recommendation matters most.

We track a Swiss outdoor sports brand whose initial set contained only branded prompts. Result: 100% visibility. Everything looked perfect. After adding industry prompts ("What is the best outdoor sports store in Switzerland?"), visibility dropped to 34%. Two-thirds of AI responses did not mention it.

A single topic. Your brand operates across multiple segments. An insurer covers home, health, auto, and pension. If all your prompts focus on the same subject, you have a blind spot on the rest. Organize your prompts by theme (in Repliq, this is the concept of topics) so you can filter your reports by segment.

How to Do It Right

Start by listing your 3 to 5 market segments. For each, create:

• 2 to 3 industry prompts: the questions your prospects ask without naming you
• 1 to 2 branded prompts: direct comparisons with your competitors
• 1 sentiment prompt: how AIs perceive your brand on that segment

Tag each prompt by topic. It sounds simple, but it changes everything when you analyze the data: instead of a single global number, you can filter by segment and see exactly where you are strong and where you are invisible.

In Repliq, the Config > Prompts page lets you create and organize your prompts by topic. The AI can also generate prompt suggestions based on an analysis of your website.

Mistake 2: Not Knowing What You Are Measuring

The second mistake we see often: starting to track without having defined what "good" means for your brand.

In AI visibility, there are two very different objectives:

Mention-first: does your brand appear in the responses? This is what most brands care about. The key metrics are visibility (% of responses that mention you), share of voice (your share vs competitors), and position (what rank you appear at in the response).

Citation-first: are your contents used as a source? This is what publishers, media outlets, and review sites care about. A specialized magazine wants to know if its articles are cited by Perplexity when someone asks "best laptop for video editing." The key metric here is the number of citations and the citation share.

Why This Changes Everything

Take a Swiss real estate agency we track. Its visibility on industry prompts is 47%: in fewer than one out of two responses, AIs recommend it when someone searches for a real estate agent in its region. This is a mention-first strategy. The metrics to watch in its Repliq dashboard: the Overview page (overall visibility and trend), the Prompts page (which prompts work, which do not), and the competitive leaderboard (who is ahead).

For a media outlet, the same dashboard pages would be viewed with a different focus: the Sources page to see which URLs are cited, and the owned citation share to measure the share of its own content in AI responses.

Define your objective before looking at the data. It seems obvious, but many brands get lost in the numbers because they have not decided what they are looking for.

Mistake 3: Looking at Global Numbers Instead of Segmenting

When you receive your first AI visibility report, the temptation is to look at the number at the top: "We are at 65%, is that good or not?"

The problem: that number means nothing on its own.

Segmentation Changes Everything

We track a Swiss watchmaker whose overall visibility is 86%. Reassuring. But when segmenting:

• On the "gift ideas" topic: 64% (AIs do not associate it with the gift context)
• On Perplexity in French: 8% (nearly invisible)
• On ChatGPT Web in Italian: 0%

The 86% was hiding a 0%. And that 0% represents a real opportunity, because it is a segment (Italian language, gift context) where the brand should be present but is not.

How to Segment Effectively

The most revealing segmentation dimensions:

• By topic: visibility on "home insurance" is not the same as on "health insurance"
• By platform: each AI has its own biases (we have measured up to 67-point gaps between two platforms for the same brand)
• By language: a Swiss insurer we track goes from 70% in French to 62% in German on the same queries
• By prompt type: the branded/industry gap is often the most revealing

In Repliq, the filter bar at the top of each dashboard page enables these cross-analyses. The Evolution page shows trends by segment over time. Resist the temptation of a single number: the value lies in the cross-analyses.

Mistake 4: Applying Generic Playbooks

The fourth mistake is believing that what works in one industry works everywhere. AI visibility dynamics are specific to each industry.

Each Sector Has Its Own Rules

In luxury watchmaking, the volume of online content is enormous: specialized blogs, collector forums, YouTube comparisons. Competition for citations is intense. A brand can have 93% visibility because the sector is saturated with content that mentions it.

In Swiss insurance, the opposite is true. Comparison sites (comparis.ch, bonus.ch) dominate the citations. If your brand does not appear in their rankings, AIs cannot find you. The strategy is not to produce more content, but to ensure that the right sources reference you.

In real estate, AI responses vary significantly by location. An agency can be highly visible in Geneva and invisible in Lausanne, simply because the local sources are different.

What This Means

Do not compare your visibility with a brand in a different industry. A 50% in cybersecurity (little content, high barrier to entry) can be worth an 85% in retail (abundant content, low differentiation).

In Repliq, automated insights take your sector into account. They identify opportunities specific to your market: missing sources, competitors outranking you on a specific topic, underserved languages.

To go further on industry differences, see our sector analyses: luxury watches, car insurance, real estate.

Mistake 5: Having the Data but Not Knowing Where to Start

Last mistake, and probably the most frustrating: you have all the data, you can see the gaps, but you do not know which action to prioritize. And the first reaction is often to want to redesign your entire website. That is almost never the right priority.

The Prioritization Framework

Here is the order in which to act, from simplest to most involved:

1. Target the third-party sources that matter. Look at which sites are cited by AIs on your key prompts. If your competitors appear on those sources and you do not, that is your contact list for PR or partnerships. This is the most cost-effective action: you do not touch your website, and the impact is direct.

In Repliq, the Sources page displays the most influential domains with an influence score (0-100). "Source opportunity" insights automatically identify sources that cite your competitors but not you.

For example, for a Swiss watchmaker we track, media outlets like elle.de, frandroid.com, or lesnumeriques.com mention its competitors in their watchmaking articles, but never mention it. A PR article in one of these outlets could directly improve its visibility on Perplexity and ChatGPT Web.

2. Correct inaccurate information. If you already appear in some third-party sources, check that the information is correct. We have seen cases where a comparison site stated that a brand did not offer a service when it has been offering it for months. That inaccurate information ends up in AI responses. A simple correction request to the site can fix the problem.

3. Optimize your key pages. When you move to your own site, do not do everything at once. Focus on the pages that AIs consult most often: homepage, main product/service page, about page. Make them clear, complete, and easy for an AI to parse.

4. Create guide-style content. The content that performs best for AI visibility is the objective, comprehensive guide. "How to choose the right home insurance in Switzerland", with your strengths and differentiators presented factually. AIs love this type of content because it is exactly what they look for when answering a question.

Conclusion

The 5 mistakes come down to one idea: measuring without method. Poorly calibrated prompts, no definition of success, overly aggregated data, generic playbooks, no prioritization. Each one is avoidable from day one.

The good news: if you are here, you have already taken the most important step. You have understood that AI visibility can be measured. The rest is method.