The buyer you wanted may have ruled you out without ever visiting your website.
They have a real requirement and an approved budget. Instead of opening ten browser tabs, they ask an AI assistant which firms could solve the problem. A few names appear, each with a short description. The buyer investigates those firms and ignores the rest.
If your business is missing, there is no rejection to analyse. No tender is lost. No salesperson hears an objection. The opportunity disappears before it reaches your pipeline, leaving the leadership team to explain a quieter quarter with incomplete information.
Most UK mid-market businesses have not checked how they appear at this stage of the buying process. A useful first assessment can be completed in two working days.
The shortlist is being formed earlier
AI adoption is not the most important change. The more consequential shift is that AI-mediated discovery now happens before many suppliers know a purchase is under consideration.
Forrester's 2026 buyer research, based on nearly 18,000 global business buyers, found that 94% used AI during the buying process. Its published analysis also reports that 47% used AI to help build the internal business case. That means a model can influence both the list of suppliers and the criteria against which they will later be judged.
Bain found a similar pattern. In its September 2025 US consumer survey, 44% of online buyers said they mainly began their journey in a large language model or divided their search between AI tools and conventional search. Bain's qualitative B2B research found buyers in small and medium-sized companies building vendor shortlists inside LLMs, then moving to websites, review platforms and video to validate the recommendations.
This behaviour is not confined to a small group of early adopters. Ofcom reported in April 2026 that 54% of UK adults used tools such as ChatGPT, Copilot or Gemini, rising to 79% among 16 to 24-year-olds.
The source of the answer matters as much as the answer itself. Bain cites ScrunchAI analysis of about 500 million citations. It found that third-party sources fulfilled 89% of unbranded prompts, meaning prompts that did not mention a company by name. Trade publications, industry directories, analyst commentary, review platforms and other external sources carry much of the evidence. A company's homepage, blog and paid advertising carry far less weight than many management teams assume.
This creates a mismatch. Marketing budgets are often built around the channels the business can see and measure, while an increasingly important part of discovery is taking place in sources it neither owns nor monitors.
Why mid-market firms are especially exposed
Well-known consumer brands have an enormous public record. Specialist B2B firms do not. In a narrow category, the available evidence may amount to a trade feature, an association listing, a handful of reviews, an old partner page and the company's own website.
That creates an opportunity. A mid-market firm can realistically become the cited name in its category when the field is narrow and the public evidence is thin. It also creates a risk. One outdated article or inaccurate directory entry can become disproportionately influential.
The problem is most acute after the business has changed. Perhaps it has moved upmarket, withdrawn from a sector, acquired a new capability, closed a division or repositioned under a new brand. If external sources still describe the old business, a model may do the same. The buyer sees an apparently sensible answer, decides the fit is wrong and moves on.
This is not simply a search problem. It is a test of whether the market's public record reflects the company the leadership team believes it now runs.
Run the Shortlist Test
The Shortlist Test has five stages. One capable person should be able to run the baseline in two days, followed by a one-hour leadership review. It does not require specialist software or an agency.
1. Build twelve realistic buyer prompts
Start with the buyer's problem, not your company name.
Your twelve prompts should span the journey from understanding the problem to choosing and checking a supplier. Include the constraints that shape a real purchase, such as sector, location, business size, accreditation, timescale and budget. Use the language customers use in meetings, even when it is less precise than your internal terminology.
For an engineering consultancy, the list might include:
- "Which UK firms can design a new food-manufacturing production line in the Midlands?"
- "What should a mid-sized manufacturer budget for an automation feasibility study?"
- "Which engineering consultancies specialise in food and beverage production?"
- "What are the alternatives to a full factory automation programme?"
Do not name your business or its competitors in the discovery prompts. You are testing whether the market evidence leads the model to you without being directed there.
Before running the test, ask two customers who have bought the service to review the list. Their corrections are useful. A leadership team often knows how it sells; customers know how they search.
2. Reduce personalisation and run the prompts consistently
Use logged-out or fresh sessions wherever the platform allows it. Record the model, date, location and relevant settings. This will not remove every source of variation, but it reduces the risk that your own history steers the answer towards your company.
Run every prompt three times on each selected platform. Repetition matters because the wording and order of recommendations can change. A firm named once is in a weaker position than one that appears reliably.
Choose the environments your buyers are likely to use. For many UK B2B firms, that will mean a sensible combination of ChatGPT, Google's AI results, Copilot and Gemini. Perplexity may also be relevant in technical markets. The aim is a disciplined baseline, not exhaustive coverage of every model.
3. Keep the evidence, not just the score
For every result, capture four things:
- whether your company appeared;
- where it appeared and how confidently it was described;
- which competitors were recommended; and
- which sources supported the answer.
Save the full transcript. A summary such as "we appeared in seven of twelve prompts" is useful, but it cannot show the board whether the business was recommended strongly, added as an afterthought or surrounded by caveats. The exact words often reveal the real problem.
The competitor and citation data can be as useful as your own result. It shows which firms the models associate with particular buyer needs and which publications, directories or review platforms are creating that association.
4. Audit every factual description
Where your company is named, test each claim against the current business.
Look for old services, former directors, historic locations, outdated headcount, sectors the firm no longer serves and positioning that predates the last strategic change. Record the likely source of each mistake. An inaccurate answer is usually inherited from an inaccurate or stale public record, which gives the team somewhere concrete to start.
Do not treat all errors equally. A minor headcount discrepancy matters less than being placed in the wrong market or described as too small for the work. Score each error according to whether it could deter the target buyer.
5. Classify the result
Place every prompt into one of four boxes.
Absent. The business did not appear in any of the three runs. For this buyer need, it is not entering the model's consideration set.
Miscast. The business appeared, but the description would discourage the customer you want. The model has the wrong view of your size, sector, services or suitability.
Named but generic. The business was included, but the answer gave no compelling reason to choose it. You are visible, but undifferentiated.
Named and accurate. The business appeared consistently with a clear, current and relevant reason for consideration.
Once all twelve prompts have been classified, calculate two headline figures for the board:
- How many prompts named the business?
- How many named it accurately?
Keep the distribution across all four boxes as the working baseline. It explains what type of intervention is required and gives the next review something meaningful to compare.
Match the response to the diagnosis
The wrong response wastes money. Each result points to a different problem.
Mostly absent: build credible third-party evidence
If the business rarely appears, publishing more generic articles on its own website is unlikely to be enough. The priority is to become present in sources that models already use for the category.
For a UK mid-market business, that may include trade publications, industry-body directories, procurement frameworks, approved-supplier lists, credible sector review platforms, awards shortlists and independent articles that quote the company by name. The precise list should come from the citations collected during the test.
This is earned visibility work. It takes longer than buying clicks, but it leaves a more durable public record.
Mostly miscast: correct the record
Trace the wrong claims back to their sources and work through them systematically. Review the LinkedIn company page, trade and association profiles, partner pages, directory entries, old press coverage and any public corporate information that a model may use.
Some records cannot be changed immediately, and legitimate historical reporting should not be rewritten. The aim is to make current, authoritative information clear enough that the old description no longer stands alone.
This is often the quickest issue to improve because it is specific. Every inaccurate source can have an owner and a target date.
Mostly named but generic: settle the positioning
If models can find the company but cannot explain why it is distinctive, the problem sits above content production. The public record is vague because the positioning is vague.
The leadership team must decide which market it wants to be considered for, which buyers it serves best and what evidence supports that claim. This is where a go-to-market and positioning review is more useful than another content plan.
Named and accurate: strengthen the validation journey
A good result still needs maintenance. Check whether the cited reasons are supported by evidence on your website and whether the claims remain current as the business changes. The goal is not merely to appear. It is to be represented for the right reasons.
Across all four outcomes, make owned content easier to interpret. Service pages should answer direct questions. State the sectors, locations and client profiles you serve. Give credible signals about engagement scope and price where appropriate. Keep key pages current, use clear page structures and check whether technical settings are preventing relevant crawlers from accessing important material.
Avoid the expensive shortcuts
Four mistakes are common.
First, do not buy a generative-engine-optimisation package before establishing your own baseline. Until you know whether the main issue is absence, inaccuracy or weak positioning, you cannot brief the right work or judge whether it succeeds. As we have written before about buying AI capability before defining the problem, starting with a solution is an expensive way to avoid the decision.
Second, do not assume the website needs a complete rewrite. It may need clearer evidence and direct answers, but the test may show that the main gaps sit in third-party sources.
Third, do not treat this as a routine SEO extension. There is some overlap, but the source mix, the output and the path to action are different. Search ranking alone cannot tell you how a model describes your business or which competitors it recommends.
Fourth, do not manufacture reviews or disguise incentivised coverage. The Digital Markets, Competition and Consumers Act 2024 applies new prohibitions to commercial practices from 6 April 2025. The CMA's guidance covers fake reviews, concealed incentivised reviews and misleading publication of review information. Where consumer reviews are involved, any visibility plan must stay within those rules. Apart from the legal risk, a manipulated public record destroys the credibility the exercise is meant to build.
Recommendation does not equal conversion
Being named earns a place in the buyer's next step. It does not win the work.
Bain's B2B research found that buyers used AI to build the consideration set, then checked the suggested firms through websites, review platforms and video. Forrester's analysis makes the same underlying point: AI can accelerate research, but buyers still look for evidence and reassurance before committing.
The journey therefore has two tests. Can the model find and describe you? Then, can the buyer verify what it was told?
The second test is where case studies, named expertise, testimonials, clear service information and a credible point of view matter. A model may create the visit, but a thin or outdated website can still end it.
Make it a leadership measure
This should not sit with an agency alone, or with a marketing manager who has no authority to resolve the strategic issues it exposes.
The test can force decisions about what category the business is competing in, which customers it wants and what it is prepared to claim publicly. Those are leadership questions. Marketing can run the process and co-ordinate the corrections, but it cannot settle them in isolation.
Someone at board level should own the two numbers: prompts in which the business is named, and prompts in which it is named accurately. Review them alongside pipeline and win rate as part of the quarterly commercial discussion.
A single test produces a snapshot. Repeating it every quarter produces a trend. That matters because models change, their sources change and the public record around the business continues to move.
Find out before the buyer does
The commercial danger is not a negative answer. It is an answer the company never sees.
Running the Shortlist Test brings that hidden stage of the buying journey into view. It shows whether the business is absent, inaccurately represented, indistinct or properly understood. More importantly, it tells the leadership team which problem to solve first.
If you would like an independent view of what your buyers are being told, Allington Advisors can run the Shortlist Test and report back with the transcripts, citation map and a clear diagnosis of the visibility, accuracy and positioning gaps. Get in touch to discuss the assessment.