Arrow 
Seven questions help a buyer verify a vendor: offer, fit, price, availability, location, next step and proof. Use the source checklist to separate verified facts from unknowns. These are editorial checks, not a published AI ranking formula.
How does an AI agent build a shortlist?
An AI agent builds a shortlist by turning a request into constraints, finding candidates and keeping only those it can match and verify. The constraints come from the person: a budget, a location, a deadline, a type of client. Each candidate is tested against them using whatever public information the agent can reach.
The seven signals below are Arrow AI's editorial framework for that test. They are not a published ranking algorithm. They describe the facts an agent needs in order to say "this option fits" with evidence.
What are the 7 signals an agent checks?
| # | Signal | The agent's question | What fails the test |
|---|---|---|---|
| 1 | Offer | Does this business do the thing requested? | Vague taglines with no service named |
| 2 | Fit | Is it for this type of customer? | No mention of size, sector or eligibility |
| 3 | Price | Can I compare it within the budget? | "Contact us" with no range or logic |
| 4 | Availability | Can it happen in the time required? | No hours, lead times or capacity |
| 5 | Location | Does it serve this place? | No service area or address |
| 6 | Next step | Can I book, buy or enquire? | Broken, hidden or image-only forms |
| 7 | Proof | Can I trust the claims? | Unattributed reviews, no dates, no cases |
Answering these questions makes a business easier to evaluate. It does not guarantee inclusion or a recommendation: the result also depends on the request, accessible sources and the assistant. Missing facts should remain explicit uncertainties, such as "price not listed".
Why does consistency matter as much as content?
Agents cross-check. If your website says you open at 8 a.m., your Google Business Profile says 9 a.m. and a directory says you closed last year, the agent has three contradictory facts. It may pick one, flag the uncertainty or prefer a competitor with a cleaner record.
The same applies to prices, service names and locations. Old prices on a cached page are a common cause; see why AI shows your old prices. A single, maintained source per fact and matching third-party profiles reduce that risk.
Which signal blocks a shortlist most often?
For many service businesses, price and next step are the easiest gaps to spot. Service pages often explain the expertise well and stop there. A person might call. An agent asked to "compare monthly fees" has nothing to compare.
Your own gaps may differ, which is why the agent-ready checklist starts with a test of your real pages rather than a general rule.
How can you test your own shortlist position?
Write three realistic delegated tasks for your market, including constraints. For example: "Find two accountants in Miami who handle crypto taxes and charge a fixed fee, and tell me their fees." Run each task in an assistant with agent or browsing features. Record which businesses were kept, which were dropped and the reason given.
Then compare your own pages with the seven signals. Each missing signal becomes a task: publish the price logic, add the service area, fix the form. Recheck the same task later under the same conditions.
What does a strong page look like?
A strong service page answers the seven questions in the first screen and the first few paragraphs: what the service is, who it is for, the price or price logic, availability, area, the action and one piece of dated proof. Everything else can follow. Read pricing pages AI agents can read for the signal most often missing.
How do you turn a software buying brief into a source checklist?
Start with the buyer's constraints, then attach a source to each requirement. A shortlist is a research output; it is not proof that a vendor meets every condition. Use three statuses: verified, contradicted and unknown. Keep unknown separate from no. This is an Arrow editorial method, not a claim about a model's private selection rules.
Consider a fictional team choosing a CRM for twelve users. It needs shared email, CSV export and a monthly budget below $600 before taxes. The example below is a blank decision framework, not a product recommendation or a measured customer result.
| Requirement | Source to inspect | What to record | Decision rule |
|---|---|---|---|
| Twelve seats below the budget | Current pricing and billing terms | Currency, billing period, seat minimum, required tier, retrieval date | Compare the actual team cost; flag missing fees |
| Shared email workflow | Vendor documentation and a trial | Supported provider, sync limits and a tested workflow | A logo on an integrations page is not a completed test |
| Usable export | Export documentation and trial output | Included objects, file format, permissions and any exclusions | Keep the requirement unknown until the needed data can be exported |
| Implementation deadline | Onboarding terms or written confirmation | Dependencies, owner and earliest realistic start | Do not substitute a sales estimate for a commitment |
| Evidence for the use case | Attributed case study with method | Customer context, date, baseline and limitations | Separate an example from a promise for this buyer |
What should a repeatable shortlist prompt include?
Use a prompt such as: Find three CRM candidates for a twelve-person team with shared email and CSV export, below $600 per month before taxes. For every candidate, show the current source URL and plan behind each requirement. Mark unverified requirements unknown. Separate recurring fees from implementation fees. Do not infer a capability from a marketing slogan.
Save the exact prompt, date, tool and mode, market, full answer and displayed source URLs. Open each source yourself. If the answer cites a pricing page but the relevant feature appears only in a different tier, correct the table before making a decision. Run a small trial with non-sensitive sample data before committing.
Repeat the same brief after changing your own source pages. Record observed inclusion, factual accuracy and citations separately. A different shortlist on the next run is an observation; it does not establish that the page edit caused the change. Avoid presenting one assistant response as your market rank.
Where does the shortlist fit in the buying journey?
Start with how buyers use AI before purchasing software, then use the public-proof guide to verify the claims behind each option. The Before they buy, they ask AI collection connects discovery, shortlisting and verification.
For a short explanation of the distinction, see brand mentions, citations and recommendations. A mention may help someone discover you; it does not establish fit or a sale.
Sources and editorial scope
Reviewed October 5, 2026. Platform-specific statements link to the documentation beside the claim. The workflows and examples are Arrow AI editorial guidance. Agent features change quickly and vary by plan and country; check the linked documentation before acting. Fictional scenarios illustrate a method; they are not customer results, measured demand or evidence of ranking gains.
Part of Before they buy, they ask AI. Explore Arrow GEO or request a free audit to see which facts an AI agent can already find about your business.