ChatGPT, Perplexity or Gemini do not build a supplier shortlist like a directory.
AI starts from a question, looks for a readable category, identifies choice criteria, mobilizes available proof and suggests options it can justify.
In B2B buying, this is decisive: your brand can be known, cited or well indexed, but remain absent from the shortlist if AI does not know why it should select you.
Short answer
AI usually builds a supplier shortlist by combining six signals:
| Signal | AI’s implicit question | Risk for the brand |
|---|---|---|
| Intent | Which problem does the buyer want to solve? | Being visible for the wrong use case |
| Category | Which family of options should be searched? | Being placed in a category that is too generic |
| Criteria | Which criteria make an option relevant? | Being compared on price, awareness or availability |
| Proof | What makes the recommendation defensible? | Being cited without proof reuse |
| Sources | Which sources corroborate the comparison? | Depending on a fragile or unfavorable source |
| Constraints | Which limits exclude some options? | Being excluded by a criterion you have not documented |
So the issue is not only: “Are we cited?”
The issue becomes:
Are we easy to select when AI has to justify a supplier recommendation?
Criteria change by market
AI does not recommend every company with the same grid. The shortlist logic changes depending on the profession, buying frequency, risk level, location and available sources.
For a recurring local service, such as a hairdresser, restaurant or landscaper, AI may give strong weight to the Google Business Profile, average rating, review volume, review freshness, proximity and sentiment in customer reviews.
For an intellectual profession in the same city, such as an accountant, lawyer or specialized consultant, the grid can be different. AI is more likely to inspect specializations, client types, local presence, covered sectors, proof of expertise and website clarity.
| Market type | Criteria often visible to AI | Risk if poorly documented |
|---|---|---|
| Recurring local service | Proximity, rating, reviews, availability, customer sentiment | Being absent or poorly ranked despite strong expertise |
| Local intellectual profession | Specialization, seniority, client type, sector expertise | Being compared as a generic provider |
| Complex B2B purchase | Choice criteria, proof, use cases, integration, risks | Being cited without entering the shortlist |
| Technical or regulated offer | Standards, compliance, thresholds, proof, usage limits | Being compared with less robust options |
| Premium offer | Avoided risk, cost of a bad decision, rejection criteria | Being reduced to a more expensive option |
GEO work is therefore not only about being cited by ChatGPT. It is about understanding the criteria grid AI applies to your market, then making those criteria visible, verifiable and easy to reuse.
What Beyond Mentions data shows
In the first consolidated wave of the Beyond Mentions Observatory, we analyzed 4,320 Perplexity sonar answers over 3 UTC days, with 6 passes per question per day.
These figures do not describe all AI systems or all markets. They describe a controlled corpus that helps observe how AI answers mobilize sources, proof, categories and shortlist logic.
| Observed signal | Volume in the corpus | Useful reading |
|---|---|---|
| Source dependency | 4,137/4,320 (95.8%) | Answers depend heavily on reusable sources |
| Proof reuse | 3,714/4,320 (86.0%) | Documented proof is often reused in justifications |
| Documentation and proof | 2,687/4,320 (62.2%) | Documentation acts as decision infrastructure |
| Shortlist and vendor evaluation | 2,168/4,320 (50.2%) | One answer in two activates supplier, shortlist or evaluation logic |
| Criteria reuse | 1,464/4,320 (33.9%) | Explicit criteria can structure the comparison |
| Category compression | 1,061/4,320 (24.6%) | A visible offer can be understood in the wrong category |
The Shortlist and vendor evaluation signal is central: in this corpus, AI answers do not only inform. They often enter selection logic.
The mechanics of an AI shortlist
1. The question sets the playing field
The same brand can be selected or excluded depending on the initial wording.
| Buyer question | Likely shortlist type |
|---|---|
| ”Which suppliers are known in this category?” | Visible players and broad categories |
| ”Which suppliers fit a high-risk environment?” | Players able to prove compliance, robustness and traceability |
| ”Which solution should we choose if budget is constrained?” | Cheaper, simpler or more standardized options |
| ”Which providers should we avoid if technical proof is weak?” | Shortlist structured by rejection criteria |
The first battle therefore happens in the questions buyers ask before contacting you.
2. AI chooses a category before choosing names
Before recommending, AI must understand which family of options to search.
If your category is clear, it can compare the right players. If it is fuzzy, it may place you in a simpler category.
| Category used by AI | Effect on the shortlist |
|---|---|
| Precise category | Competitors and criteria are more relevant |
| Category too broad | You are compared with less demanding alternatives |
| Category too technical | The buyer may not understand the value |
| Competitor category | Another market’s standards structure the decision |
This directly connects with Category Compression Risk: the wrong category changes the shortlist before your offer is even evaluated.
3. AI applies explicit or implicit criteria
A shortlist is never neutral. It always relies on a grid.
That grid can be explicit:
- compliance;
- available proof;
- integration;
- sector specialization;
- deployment capability;
- support;
- total cost;
- usage limits.
It can also be implicit:
- awareness;
- source quantity;
- message simplicity;
- availability of customer cases;
- ease of explaining the difference;
- presence in existing comparisons.
If your differentiating criteria are not published, AI uses the criteria that are easiest to find.
4. AI selects the most defensible options
In complex buying, AI does not only look for suppliers. It looks for recommendations a buyer can defend.
| Easy option to defend | Hard option to defend |
|---|---|
| Clear category | Vague positioning |
| Named criteria | Generic claims |
| Verifiable proof | Unsourced promises |
| Concrete use cases | Feature list |
| Explicit usage limits | Overbroad recommendation |
| Corroborated sources | Dependency on a single page |
The shortlist therefore often favors brands whose documentation reduces the risk of a poor recommendation.
Beyond Mentions panels confirm the framing effect
In the brief Beyond Traffic: measuring Decision Presence before clicks, the same signals change strongly depending on the type of question asked.
| Public panel | Shortlist / vendor evaluation | Reading |
|---|---|---|
| Market baseline | 194/1,152 (16.8%) | Broad questions activate supplier selection less often |
| Conceptual boundaries | 871/1,152 (75.6%) | Framing questions strongly trigger comparison |
| Launch framing | 817/1,152 (70.9%) | Need wording orients the shortlist very early |
| Category compression | 615/864 (71.2%) | Wrong-category tests massively activate selection logic |
The practical conclusion: to understand your AI presence, testing your brand name is not enough. You must test the questions that build the shortlist before the first conversation.
Why a supplier is excluded
Absence from the shortlist can come from several different problems.
| Symptom | Probable cause | Correction |
|---|---|---|
| Your brand is cited elsewhere but absent from the shortlist | Presence without preference | Connect the brand to selection criteria |
| You appear in the wrong family of players | Weak category fit | Clarify category, use cases and wrong substitutes |
| AI recommends more generic competitors | Your criteria are implicit | Publish matrices, thresholds and rejection criteria |
| Your proof is not reused | Low proof reuse | Turn cases, figures and standards into extractable blocks |
| AI describes you as expensive | Value is not tied to risk avoided | Document the cost of a poor decision |
| A third-party source describes you poorly | Unfavorable source dependency | Create more precise and corroborated sources |
This avoids a common mistake: assuming absence only comes from lack of visibility. Often, the problem is lack of justification.
What to publish to enter the right shortlist
A useful shortlist page does not only sell the offer. It helps the buyer choose correctly.
| Block to publish | What AI can do with it |
|---|---|
| Category definition | Understand which family to search |
| Choice criteria | Compare options with an explicit grid |
| Supplier matrix | Differentiate types of players |
| Verifiable proof | Justify a recommendation |
| Use cases | Connect the offer to a buying situation |
| Usage limits | Avoid overbroad recommendation |
| Rejection criteria | Exclude insufficiently documented options |
| Buyer FAQ | Prepare first-conversation questions |
Format matters as much as substance: clear sentence, named criterion, associated proof, threshold or limit whenever possible.
Example of an extractable block
A weak block looks like this:
“Our solution is robust, expert and suited to demanding companies.”
A more useful block for an AI shortlist looks like this:
“For a high-constraint environment, a supplier should document the applicable standard, proof of compliance, criticality threshold, exposure scenario and cases where the offer should be rejected.”
The second formulation helps AI compare, justify and sometimes exclude.
How to audit your AI shortlist
Build a corpus of buyer questions, then code the answers.
| Step | Control question |
|---|---|
| Category | Which category does AI place us in? |
| Competitors | Who are we compared with? |
| Criteria | Which criteria structure the shortlist? |
| Proof | Which proof is reused? |
| Exclusions | Which criteria eliminate options? |
| Position | Are we favorite, alternative, absent or simple source? |
Repeat the questions several times. One answer can vary. A repeated pattern indicates a more stable cognitive map.
Key takeaway
AI builds a supplier shortlist from what it can understand and justify.
A B2B brand increases its chances of being selected when it makes visible:
- the right category;
- the right criteria;
- verifiable proof;
- relevant use cases;
- usage limits;
- rejection criteria;
- sources that corroborate its value.
The real question is not:
“Does AI know our brand?”
The real question is:
“Does AI know why our brand should be selected in this buying situation?”
Read next
- How do ChatGPT, Perplexity and LLMs influence B2B buying?: understand when the shortlist starts forming.
- How do ChatGPT and LLMs recommend a company?: see why criteria change by market.
- Why am I cited by ChatGPT or Perplexity but not converting?: distinguish citation, preference and recommendation.
- Stop mourning clicks: control the AI-generated shortlist: measure Shortlist inclusion rate.
- Documentation Blind Spot: turn implicit proof into reusable blocks.
FAQ
Is an AI shortlist reliable?
It can be useful to explore a market, but it must be checked. An AI shortlist depends on the buyer question, available sources, the category used and the proof AI can reuse.
Why do some suppliers appear more often than others?
They often appear because their category is clear, their proof is published, their use cases are easy to understand and several sources make the recommendation defensible.
How do you get into an AI shortlist?
Document buyer questions, choice criteria, verifiable proof, usage limits, relevant use cases and the situations where the offer should be selected or rejected.
Which KPI should measure the AI shortlist?
The main KPI is Shortlist inclusion rate, read together with criteria reuse, Proof reuse, category fit and Decision Share of Voice.
What question does the buyer ask AI?
Which documentation simplification can lower the standard?
Which technical requirement must be clearly formulated?
Which evidence should be requested or published?
Which criterion excludes an insufficient answer?