Your prompt set is well built, your sampling is honest, and the answer keeps coming back the same: you are absent. Before you spend another quarter on retrievability work, it is worth checking whether the problem is that you are competing for a category label your buyers do not use, because that is a positioning failure and no amount of measurement will fix it.
The category term your buyers actually use
Assistants answer the question asked. When somebody types a request for the best option in a category, the model assembles an answer around that category as it understands it, drawing on how the wider web describes and groups things. If your company describes itself with a term that sits outside that grouping, you are not losing the comparison. You are not in it.
Marketing teams create this gap honestly. A company invents a phrase to describe what it does because the existing labels felt reductive, the phrase becomes internal shorthand, it propagates through the site and the deck and the job listings, and eventually nobody in the building says the plain words a buyer would say. Meanwhile the buyer types the plain words.
Three symptoms suggest a category problem rather than a visibility problem:
- You appear when the prompt names you, and never when it does not.
- Your competitors in the answers are companies your sales team does not recognise, which means the model has grouped the category differently from how you have.
- Sales calls open with the prospect describing what you do in words that appear nowhere on your website.
The third one is the strongest signal and the cheapest to collect. Ask three account executives what customers call the thing you sell. If their answer differs from your homepage, you have found something.
Testing three framings against the same prompts
The test is a controlled comparison. Hold the buyer intent constant and vary only the category noun, then read which framing surfaces you.
Step 1. Name three candidate framings.
- A: the term you currently use. Whatever is on your homepage.
- B: the plain-language term. What a buyer says out loud, usually the older and less exciting label.
- C: the adjacent framing. The neighbouring category you could credibly be grouped into, often defined by the job the buyer is doing rather than the product type.
Step 2. Write the same prompt shapes for each framing. Use at least four intent shapes per framing so you are not reading a single phrasing artefact:
| Intent shape | Template |
|---|---|
| Best-of | "What are the best [framing] for [buyer type]?" |
| Constrained best-of | "Best [framing] for a [size or segment] team with [constraint]" |
| Alternatives | "What are the alternatives to [a well-known option in that framing]?" |
| Problem-first | "How do I solve [the underlying problem]?" |
| Comparison | "Compare the main [framing] options for [use case]" |
That is 15 prompts if you run five shapes across three framings. Add regional or segment variants only if they matter commercially, because every variant multiplies the run cost.
Step 3. Run it properly. Same models, same versions, same day, multiple samples per prompt, recorded verbatim. One run of one prompt tells you nothing, because these systems vary between identical requests. If you have not fixed your sampling approach yet, do that before you interpret anything here. The prompt-set templates give you a structure to fill in.
Step 4. Record the same fields for every response. At minimum: whether you were named, in what position, in what role (recommended, listed, mentioned in passing, named as an alternative), which competitors appeared, and which sources were cited.
The comparison is only valid if the framing noun is the sole thing that changed. Rewriting the prompt to be more flattering under framing B is the easiest way to fool yourself, and it is the most common error in this exercise.
Reading which framing you already own
Aggregate by framing, not by prompt, and look at four things.
| What to read | What it suggests |
|---|---|
| Presence rate by framing | Which category the model already associates you with |
| Competitor set by framing | Whether each framing describes a market you actually compete in |
| Cited sources by framing | Which publications and directories define each category, and whether you appear in them |
| Role when present | Being listed as an alternative is different from being recommended, and the gap is usually a credibility gap |
Four common patterns:
- Absent under A, present under B. The market already files you under the plain term. Your language is the outlier. This is the cheapest situation to fix.
- Present under A only. You have some ownership of a term nobody searches with. Useful for defence, insufficient for demand.
- Absent under all three. This is not a framing problem. It is an evidence problem, and the work is entity clarity and third-party coverage.
- Present under C with unfamiliar competitors. Worth a serious conversation, because the model may be describing a real adjacent market you have not been treating as competitive.
One caution on interpretation. Presence under a framing is a statement about how the current web describes the category, sampled on a date, against particular model versions. It is not a market-size estimate and it is not a forecast. Report it as what it is and re-run it before making a large decision on the strength of a three-month-old result. The wider method for turning these runs into market reading is in using AI answers as positioning research.
Here is an illustrative example, offered only to show the shape of a result and not as a measured finding: a company selling scheduling software to home-services firms describes itself as a "field operations platform" (A), while buyers say "scheduling software for contractors" (B), and an adjacent framing would be "job management software" (C). A run of this test might show the company absent under A, present under B alongside competitors its sales team knows well, and present under C against a larger and unfamiliar set. That result would point to a language problem under A and a strategic question under C.
What changing category framing costs you elsewhere
This is the section teams skip, and skipping it is how a measurement exercise turns into an expensive mistake. Category language is not a metadata field. It is load-bearing across the business.
- Sales. Reps have a story and a set of objections that fit the current framing. Changing the noun changes the discovery conversation and the competitive set they are compared against.
- Pricing power. Categories carry price expectations. Moving to a plainer, more commoditised label can pull your price anchor down, which is often the real reason a team invented the fancier term.
- Product roadmap. Claiming a category creates feature expectations. Buyers will ask for the things that category is assumed to include, and "we do not do that" is a harder answer once you have claimed the label.
- Hiring and investors. Both audiences respond to category signals, sometimes on a different timetable from your buyers.
- Existing search equity. Pages, links, and third-party listings are attached to the current framing, and a rewrite discards some of that.
- Internal cost. Site copy, collateral, help documentation, integrations directories, and every third-party profile.
The point is not that changing is wrong. It is that this is a company decision with a real bill attached, and it cannot be made by whoever owns the AI visibility report. Treating positioning as a technical fix is how you end up with a homepage the sales team refuses to use.
Deciding: adapt the framing or contest the term
Three defensible outcomes, and one you should refuse.
Adopt the plain term. Choose this when the test shows the market already files you under B, the plain term describes what you genuinely sell, and the pricing implications are acceptable. Cheapest path, fastest to show a change, least strategic risk. Implementation is largely evidence work: consistent naming, accurate descriptions on the surfaces that get cited, and the entity groundwork described in entity clarity.
Run both. Use the plain term where buyers meet you and keep your own term for the story you tell once you are in the room. Common and workable. The failure mode is inconsistency severe enough that neither term accumulates evidence, so decide explicitly which surfaces carry which language.
Contest the term. Choose this only when you have the resources and the time horizon to make a category label mean something on the wider web, which takes sustained third-party coverage, not a page rewrite. Category creation is a multi-year programme with a high failure rate, and the visibility measurement will look flat for a long time before it moves.
Refuse the fourth option. Do not claim a category you cannot substantiate. If the test shows a category with attractive demand and you do not have the product, the customers, or the references to belong there, describing yourself that way produces demos that go nowhere, churn, and third-party material that eventually contradicts you. Assistants assemble answers from what the web says about you, and a claim your evidence does not support tends to get corrected in public. Category framing is a description of something true, tested for whether buyers recognise it. It is never an aspiration published as a fact.
Whichever route you take, re-run the same prompt set on the same schedule afterwards. That is the only way to know whether the framing decision moved anything, and it belongs in the ongoing programme described on competitive intelligence.
In Plain English
If assistants never name you, check whether you are competing under a category label your buyers do not use before you spend money on anything else. Run the same buyer prompts across three framings, your current term, the plain-language term, and a credible adjacent one, then read which framing already surfaces you and against which competitors. This helps founders, heads of marketing, and strategists who have a clean measurement programme and a stubbornly empty result. Use it before committing to a long visibility programme. Remember that changing category language has costs across sales, pricing, and roadmap, that the decision is not the analyst's to make alone, and that claiming a category you cannot substantiate is worse than being invisible in it.
Get the prompt-set template pack to run the three-framing test with a structure you can hand to a colleague.