Is your brand known for the attributes you want to own? A positioning worksheet

To find out whether your brand is known for the attributes you want to own, ask AI engines two different questions and keep the answers apart: what is your brand known for, and which brands are known for each quality. Sentiment cannot answer either one, because an answer can be warm about your brand and still never connect it to the quality you sell. The method below compares your intended positioning with the associations engines actually state, one attribute at a time, and records what you would need to check before acting.

In short

• Sentiment tells you how an answer feels about a brand. An attribute tells you which quality the answer links to it. Track them separately.
• Ask two questions per quality: does the engine list it when describing your brand, and does it name your brand when asked who is known for it?
• Those two scores can disagree, and the disagreement is the most useful finding.
• An attribute can be good or bad. "Expensive" is an attribute in the same way "Ease of use" is.
• Everything here is an observation of engine answers. It does not show why an engine said it, so treat each gap as a hypothesis.

Why is positive sentiment not the same as positioning?

Positive sentiment means an answer spoke well of your brand; positioning means the answer tied your brand to a specific quality you want to own. The two come apart often. An engine can recommend you warmly for reasons you never chose, or describe you accurately in every respect and still name a competitor first when a buyer asks who is best at the thing you care about.

Two neighbouring measurements help, but neither is this one. Sentiment trends show how an answer frames a brand, positive or negative, top recommendation or warned against. Brand reasons count the strengths and weaknesses an answer gave for a brand in your tracked prompts. Sentiment is about how an answer frames a brand, and reasons are about why engines favour or caution against it. Attributes are about what a brand is known for, good or bad. For the sentiment side of this argument, see AI brand sentiment: read the evidence behind a positive score.

What does "known for" actually mean in an AI answer?

In this post an attribute is a short, neutral quality phrase such as "Ease of use" or "Pricing transparency" that an engine's answer associates with a brand. An association is how prominently one brand's own description lists that quality. Market prominence is how prominently a brand appears when the engine is asked which brands are known for that quality across a market.

These come from two fixed questions in DiscoveredBy's attribute studies (see the Attributes documentation). The first asks what a brand is best known for, listing qualities from most to least prominent, good or bad. The second asks, of a market, which brands are best known for one named quality. The second question does not name your brand, so the engine is not steered toward it.

That distinction matters for how you read the results. The first question shows how you are described. The second shows whether you are the answer when the quality is the question. Buyers ask both kinds of thing.

Enlarge image

What does the market question add that a description does not?

The market question shows whether you are named at all when a buyer asks for the quality, and who is named ahead of you. That is a competitive reading that a description of your own brand cannot give, because in a description you compete with nobody.

In a DiscoveredBy study the selected brand's market prominence is averaged over the engines that answered, counting zero for an engine that answered without naming you. Position is one plus the number of brands with higher prominence, so "#3 of 7 named" means two brands are ahead (brands with equal prominence share a position). The leader is the brand with the highest prominence and can be one you do not track.

The wording of your market changes what all this means. A broad market such as "software" puts you among every well-known software brand, where a small company is rarely named. A narrower one changes the field you are ranked in. Set it deliberately and keep it stable, because studies are only compared with earlier studies that asked the same market in the same language. If your market is empty, the Attributes screen shows a suggestion from your site research, and a study that starts with no market saves that suggestion as your market if it names none of your tracked brands. Check it before the first study runs.

How do you read association and market prominence together?

Read them as a pair, per attribute, and never subtract one from the other. The docs state plainly that the two scores are not comparable with each other: one comes from a question about a single brand, the other from a question where every named brand competes for the same ranks.

Each cell is a place to start looking, not a diagnosis. Neither score says what caused the engines to write what they wrote.

How do you set up an attribute study?

Decide your target attributes first, then let the engines say what they think. In the product this means setting a market, adding the attributes you want to be known for, and reading the latest study.

Owners and editors can add up to 10 of their own attributes, each a label of up to 80 characters without any brand names. Every one of them is asked in the market question (which needs a market to be set; without one, the study skips that question), and each is listed on the screen even when no engine associated it with your brand, which is how you see a zero rather than a missing row. A study also asks about up to four of your brand's strongest other attributes and up to eight strong attributes of any brand in the study, which is how a quality only a competitor owns can appear. In all, a study asks at most 18.

Studies run automatically once a week for eligible projects, or on demand with Run now, which is refused while another study runs and for 24 hours after the previous one. Availability depends on your plan; see plans and limits. Studies ask the chat engines on your plan plus ChatGPT (app), but not Google AI Overviews, Google AI Mode or Gemini (app), so your worksheet should record which engines answered. A study is separate from your tracked prompts and changes no visibility metric.

Enlarge image

The positioning worksheet

Fill in one row per intended attribute before you open the results, so the results cannot rewrite your intent. Copy this into a spreadsheet or a document.

POSITIONING WORKSHEET

Study date:                 Market asked (exact wording):
Language:                   Engines that answered:
Compared with study dated:  (same market, language, question versions?)

STEP 1. INTENT (fill before opening results)
Attribute label (no brand names) | Why it matters to buyers | Evidence we can back it with

STEP 2. OBSERVATION (one line per attribute)
Attribute:
  Association score (0-100):            Engines that named it (x of y):
  Market prominence (0-100):            Named by (x of y):
  Position (#n of N named):             Leader:            Leader tracked? Y/N
  Change since last comparable study:   New / Rising / Falling / Steady / Gone / No comparison
  Best supporting quote (association):
  Best supporting quote (market ranking):
  Sources shown for the quotes:

STEP 3. SEPARATE THE THREE READINGS
  Sentiment of the answers naming us (from Sentiment):   positive / neutral / negative / unclassified
  Association reading:                                   high / low / not named
  Market reading:                                        leader / ahead of us / not named
  Do these three agree? If not, which one is the surprise?

STEP 4. CHECKS BEFORE ACTING
  [ ] Quotes actually say what the score implies
  [ ] Attribute group is not broader or narrower than we mean it
  [ ] Only one study so far, or a repeat across weeks?
  [ ] Engine subset the same as last time?
  [ ] Market wording still the one we want to be judged in

STEP 5. DECISION
  Hypothesis (what might explain the gap, stated as a guess):
  Action type: verify a claim / add evidence to a page / earn third-party coverage / accept / drop attribute
  Owner:                     Re-check date:

Keep step 3 honest. It is the step that stops "the sentiment is positive" from standing in for "we own this quality".

Worked example

Illustrative example: Quillstone and its competitors are fictional, and the numbers are made up to show the method.

Quillstone sells document-review software to legal and compliance teams. Its market is "document review software for legal and compliance teams". Its team wants to be known for three things: Ease of use, Audit trail and Security certifications. Three engines answered every question. The competitors are Brieflane and Clausewise.

Scores follow the documented scale: 100 for a quality listed first, minus 10 for each later position, and 0 for an engine that answered without naming it.

The arithmetic: Ease of use association is (100 + 90 + 0) ÷ 3 = 63.3; its market prominence is (80 + 70 + 0) ÷ 3 = 50.0. Audit trail association is (80 + 0 + 0) ÷ 3 = 26.7; its market prominence is (100 + 100 + 90) ÷ 3 = 96.7. Security certifications is 0 in both.

Reading the rows. Audit trail is the interesting one: Quillstone leads the market ranking but engines seldom list it when describing Quillstone. That is the "low association, high prominence" case. The team's first task is to read the quotes and see which qualities the descriptions lead with instead, then decide whether the product pages state the audit trail as plainly as they state ease of use.

Ease of use is the reverse: described often, but a competitor is named ahead. The team checks the leader's supporting quotes and sources before writing anything.

Security certifications is a zero in both. Because Quillstone added it to its own list, the row exists at all. The team does not conclude that it lacks certifications, only that the engines did not connect it to Quillstone. A separate check of its own pages against the claim comes next.

Change over time. Suppose the previous comparable study, over the same three engines, gave Ease of use an association of 43.3. The change is 63.3 minus 43.3, or 20 points. The threshold for Rising or Falling is 15 points, so this reads Rising. The team logs it as one observation and waits for another study before crediting any specific edit.

Enlarge image

What can you do with a gap?

Turn each gap into a testable hypothesis, then pick the smallest action that would test it. The worksheet's action types map to different work.

• Verify. If engines link you to an attribute you do not want or that is wrong, read the quotes and sources first. A linked source shows where the engine placed its citation; it does not prove the page says what the quote says. For claims that are simply wrong, what to do when AI gets your pricing or product facts wrong covers the fix path.
• Add evidence to your own pages. If a quality is missing from descriptions, check whether your pages state it plainly with something concrete behind it.
• Earn coverage elsewhere. If a competitor leads the market ranking, look at the sources beside its quotes to see where the association appears. Whether that helps you is a hypothesis.
• Accept or drop. Some attributes are not worth owning. Removing one of your attributes only takes the mark off; the history stays.

Attributes also sit next to the negative view. What engines say your product is bad at belongs to a different study; see what AI says your product is bad at. Attributes make no judgement about direction, so an unwelcome one such as "Expensive" appears alongside the welcome ones.

Common mistakes and what this cannot tell you

• Treating the score as a ranking cause. An association or a position describes what engines said. It does not explain why they said it, and a page change is not shown to move it.
• Comparing association with prominence. The two scales come from different questions. Compare each across brands and across studies, not with each other.
• Reading one study as a trend. A single week can be noise. Compare only studies with the same market, language and question versions; a change to the market changes what is being compared.
• Trusting the grouping blindly. Attribute groups are an AI model's judgement and wording. Two similar qualities can land in separate groups, or a group can be broader than you would draw it. The quotes are the check.
• Reading a zero as "we lack this." A zero means the engines that answered did not name the quality for you. It says nothing about whether you have it. If no engine answered, the number is unavailable, not zero.
• Expecting local or persona views. Studies are not run per city or country, and use no persona; ChatGPT (app) is always asked with the United States as its country.
• Mixing spelling variants. Untracked brand names are merged on the screen by rules that can join a company and its product. Check the names in the ranking before you read a competitor's position.

Frequently asked questions

What is a brand attribute in AI search?

It is a short, neutral quality phrase, such as "Ease of use", that an AI answer associates with a brand. In DiscoveredBy, the phrase is extracted from the engine's answer to a question about the brand, along with the sentence that supports it. Attributes can be positive or negative.

How is this different from brand sentiment?

Sentiment describes how an answer frames a brand; an attribute names the quality the answer links to it. A positive answer might mention no attribute you care about. Keep them as separate columns, as the worksheet does.

Can I choose which attributes are measured?

Yes, up to 10 of your own, added by an owner or editor. Provided a market is set, they are always asked in the market question, and they are always listed. The study adds up to four of your other strongest attributes and up to eight more from any brand, so competitor-owned qualities also appear.

Why does my market wording matter?

The market phrase is included in both questions. A broad phrase puts you among every well-known brand and a narrow one changes who you compete with. Studies are compared only with earlier studies that used the same market and language, so decide the wording early and change it rarely.

How often should I review attributes?

Studies run weekly for eligible projects, but weekly reading is rarely needed. Review when you change positioning, launch a page meant to support an attribute, or start a new quarter, and prefer patterns over two or more comparable studies to a single week.

Does a high market position mean AI will recommend us?

No. Market prominence measures how prominently engines name you for one quality when asked about it. Recommendations in buying answers depend on many things, and this reading cannot show what drove any of them.

Next step

Start with your intent, not the results. Write the three to five qualities you want to own, phrased without brand names, and set a market wording you would be happy to be judged in. Then run an attribute study in DiscoveredBy and fill in the worksheet from the evidence quotes. Create or sign in to your account, and see the Brand perception feature overview for how attributes sit with sentiment, reasons and objections.