AI visibility starts with the questions you test
Marketers are used to thinking in keywords. For years, search strategies have been built around identifying the terms people type into Google, measuring rankings and then improving the pages competing for those searches. AI search changes the format of the interaction because, instead of entering a short phrase such as “accountancy firms Manchester”, someone might ask: “Which accountancy firms in Manchester would you recommend for a fast-growing technology company that needs help with R&D tax credits?”
The underlying intent is similar, but the AI prompt contains far more context. This is why one of the most important parts of measuring AI visibility is building the right set of prompts. If your prompts do not reflect the questions customers actually ask, your AI visibility score will tell you very little about your real position in the market.
Keywords and prompts are not the same thing
Keywords are still valuable because they help marketers understand demand, but a keyword and an AI prompt should not be treated as interchangeable. A traditional keyword is often short and fragmented, while an AI prompt can contain a problem, company type, location, budget, requirements and purchasing criteria within a single request.
Consider a company selling cybersecurity software. A traditional SEO keyword might be “attack surface management software”, while an AI search could ask: “What are the best attack surface management platforms for a UK company with around 500 employees?” Another user might ask for alternatives to a particular competitor or describe the underlying problem without knowing the product category at all. Those questions may ultimately lead towards the same type of product, but they represent different customer needs, and a useful prompt strategy needs to capture those differences.
Do not build your benchmark around your brand name
One of the easiest mistakes to make when testing AI visibility is asking too many branded questions. If you ask ChatGPT what your company does, you are primarily testing whether the system knows about a brand it has already been given. That can still be useful for checking accuracy, sentiment and brand understanding, but it tells you very little about whether you are being discovered by potential customers.
The more commercially important questions are the ones asked before somebody knows your name. If a potential buyer asks for the best companies in your category, do you appear? If they describe the problem you solve without mentioning your product, are you recommended? If they compare two competitors, does your organisation appear as another option? These are the prompts that expose genuine visibility gaps.
Start with the customer journey
A strong prompt strategy should reflect different stages of the buying journey. At the beginning, a customer may not even know what category of product or service they need. They simply have a problem. Later, they may begin looking for providers, comparing specific companies and eventually investigating one brand in more detail before making a decision.
A recruitment consultancy, for example, might test an early-stage question about how a technology company can recruit senior software developers more quickly, followed by a discovery prompt asking which recruitment agencies specialise in technology roles in Leeds. A later comparison prompt could ask for alternatives to a major competitor, while a final-stage prompt could ask whether a particular agency is suitable for scaling technology companies. Each question measures something different, and together they create a much more realistic view of the customer journey.
Build discovery prompts
Discovery prompts are some of the most valuable questions in an AI visibility benchmark because the user is actively looking for possible solutions but has not selected any brands. These prompts often contain language such as “best”, “recommended”, “top”, “which companies” or “who should I use”, although they should still sound like realistic customer questions rather than artificial marketing tests.
For a digital marketing agency, a useful discovery prompt might ask which agencies are best suited to B2B technology companies in the UK. An accountancy firm could test which firms are recommended for a growing ecommerce business, while a software provider might ask which tools are best for monitoring third-party cyber risk. These questions matter because the AI system is effectively creating a shortlist. If your competitors repeatedly enter that shortlist and you do not, you have identified a genuine AI visibility problem.
Include problem-led prompts
Customers do not always know the name of the service they need, so they often describe the problem instead. Someone may not know they need a digital adoption platform and instead ask how to reduce the amount of training employees need when new software is introduced. A buyer may not know the term external attack surface management and simply ask how they can identify internet-facing assets their security team does not know about.
Problem-led prompts allow marketers to test whether AI systems associate their brand with the problems it actually solves. They can also highlight valuable content opportunities. If competitors appear regularly for a problem-led question and your organisation does not, there may be an opportunity to create better educational content around the problem rather than simply producing another service page.
Test recommendation prompts
Recommendation prompts sit closer to commercial intent because the user already knows roughly what they need and wants the AI system to help choose a provider. These should form a significant part of most AI visibility benchmarks, and the strongest examples usually contain realistic constraints rather than broad requests for the “best” company.
Instead of asking for the best law firms generally, a customer might ask which UK law firms are suitable for a technology startup raising its first institutional funding round. Instead of asking for the best hotels, someone might ask which hotels in Manchester are suitable for a business traveller who wants to be close to Piccadilly station and needs meeting space. Constraints make the prompt more representative of an actual buying decision and reveal whether the AI understands where your organisation fits within its category.
Test competitor comparison prompts
Competitors already occupy space in your customers' minds, so they should occupy space in your prompt strategy too. Comparison prompts can reveal whether your organisation is considered a credible alternative to established competitors. A software company might ask for the best alternatives to a competitor or ask how a competitor compares with other platforms for a specific use case.
You do not always need to include your own company in the question. In fact, leaving it out can be more revealing because if the AI independently introduces your company as an alternative, that is a stronger indication of visibility than explicitly asking it to compare you. Comparison prompts can also expose positioning gaps, showing whether your brand appears for certain use cases but disappears when particular capabilities, industries or company sizes are mentioned.
Include accuracy prompts
Being visible is not enough if the information being presented is wrong. Every AI visibility benchmark should therefore contain prompts designed to test accuracy across products, locations, services, pricing approach, target audience and areas of expertise.
The purpose is not simply to see whether the AI mentions you, but to understand whether the digital picture of the business matches reality. An AI system recommending a discontinued service, incorrect location or outdated product could create a customer experience problem even if the brand itself is highly visible. This is particularly important for businesses that have recently rebranded, changed their offering, entered new markets or removed older services.
Measure sentiment and positioning
Marketers should also understand how AI systems describe their brand. A company may be visible but consistently positioned in a way that does not support its current strategy. Questions about what a company is known for, who typically uses it, its strengths and weaknesses, or how it compares with competitors can reveal which attributes are strongly associated with the organisation and which messages are failing to travel beyond its own website.
This can be particularly valuable following a repositioning exercise. Your website may have changed six months ago, but the wider information ecosystem could still be reinforcing the previous proposition. Monitoring positioning helps marketers understand whether the market-facing version of the brand is actually being reflected in AI-generated answers.
Use customer language, not marketing language
The quality of the benchmark depends heavily on how naturally the prompts are written. Marketing teams are often too close to their own terminology, while customers may describe the same need very differently. Sales conversations, customer service tickets, Search Console queries, internal site searches, CRM notes, reviews and community discussions can all provide clues about the language customers actually use.
This is especially valuable because AI search is conversational. A customer is unlikely to carefully reproduce your positioning statement or the terminology used on your service page. They will explain what they want in their own words. The closer your prompts get to that language, the more realistic your benchmark becomes.
Build variation into your prompts
There is rarely one definitive way to ask a commercial question. One person might ask who the best SEO agencies in Manchester are, while another asks for a Manchester agency capable of increasing organic traffic. A third might explain that they run a £10 million B2B company and want an SEO partner in the North West.
Testing some variation prevents marketers from drawing conclusions based on a single phrasing. AI answers are not fixed search results, and relatively small changes in wording or context can affect the recommendations produced. Your objective is not to test every possible variation, but to create enough diversity to understand whether a visibility pattern is consistent.
Decide how many prompts you actually need
More prompts do not automatically create better measurement. A list of 500 poorly chosen prompts can be less valuable than 50 commercially meaningful questions. For most organisations beginning AI visibility monitoring, the goal should be useful coverage rather than sheer volume.
Your benchmark needs enough prompts to represent your important products, services, customer types, problems and stages of the buying journey. A smaller company with one core service may need a relatively compact set, while a national retailer with hundreds of categories will naturally require a broader framework. The important thing is that every prompt has a reason to exist. If you cannot explain what customer behaviour or commercial opportunity a prompt represents, consider whether it belongs in the benchmark.
Group prompts so the results are actionable
A prompt list becomes much more useful when it has structure. Rather than treating every result independently, marketers can group questions around meaningful themes such as discovery, recommendations, service lines, geography, industries or customer segments.
This makes patterns easier to identify. Perhaps your company performs strongly for brand questions but poorly for discovery, appears frequently for one service but is almost invisible for another, or performs well nationally but disappears when regional criteria are added. Those insights are much more useful than a single overall visibility score because they give the marketing team somewhere specific to investigate.
Track competitors consistently
One of the most valuable parts of AI visibility monitoring is understanding who appears when you do not. Do not only record whether your own brand was mentioned. Track the companies being recommended repeatedly because you may discover competitors you did not previously consider major organic search rivals.
AI systems can create a different competitive landscape because they are being asked to interpret the suitability of businesses rather than simply return pages matching keywords. If one competitor appears across a large proportion of your highest-value prompts, investigate what content they have, how third parties describe them, where they are being mentioned and which areas of expertise are consistently associated with them. Competitor visibility can often point towards the signals your own organisation is missing.
Do not judge AI visibility from one test
AI-generated answers can change, so running one prompt once and taking a screenshot is not enough to establish that a brand is either visible or invisible. The same platform may generate different recommendations when a question is repeated, while different platforms can also produce entirely different shortlists.
An AI visibility benchmark should therefore be treated as ongoing measurement rather than a one-off ranking report. Look for patterns across multiple prompts, platforms and measurement periods. The useful question is not whether ChatGPT mentioned you once, but whether your organisation consistently appears when potential customers ask commercially important questions.
Turn prompt gaps into marketing actions
The value of a prompt strategy is not the prompt list itself. It is what the results tell you to do. If a company performs well when people ask directly about the brand but rarely appears in category recommendations, that could suggest a discovery or authority problem. If it appears for broad category prompts but disappears when a particular industry is mentioned, there may be insufficient evidence connecting the company to that sector.
Competitors may also be appearing because they have stronger comparison content, more useful educational material or a larger body of authoritative third-party coverage. If the AI consistently gets factual details wrong, the priority may instead be correcting and strengthening the information available across the website and wider web. The prompt benchmark provides the diagnosis, while the wider marketing strategy determines what to change.
Revisit your prompts as the business changes
A useful prompt set should not remain frozen forever. Products change, competitors change, customer behaviour evolves, new AI platforms emerge and marketing priorities move. The benchmark should therefore be reviewed periodically to ensure it still represents the questions that matter commercially.
A new product launch should introduce new prompts, expansion into a new market may require geographic variations and a change in ideal customer profile should alter the situations being tested. The prompt benchmark should evolve alongside the organisation rather than becoming another static marketing report.
Start with the questions that could influence revenue
AI visibility can quickly become complicated if marketers try to measure everything, but it does not need to be. Start with the customer by identifying the problems they are trying to solve, the recommendations they might request, the companies they are likely to compare and the questions they ask before making a decision. Turn those moments into prompts, then test whether your brand is part of the answer.
A good AI visibility prompt strategy is not a collection of clever questions designed to make your company appear. It is a structured simulation of the conversations your future customers are already having. If your competitors keep appearing in those conversations while you do not, you have found somewhere worth focusing your marketing effort.