The way to onboard a client into AI visibility monitoring without a sprawling prompt list is to run a kickoff intake before you create anything: collect the client's business goal, the facts they approve, every name their brand goes by, and the decisions their buyers make. Then turn only those decisions into prompts, a small set you can explain line by line. The intake is also what keeps the monitoring honest later, because each prompt, alias and fact traces back to something the client said. A starter set is a hypothesis about what buyers ask, and you should expect to change it.

In short

Interview before you configure. Four things drive the setup: the business goal, approved facts, brand names and aliases, and the questions buyers ask at each decision stage.
Build the prompt list from decisions, not from keywords. Every prompt should be traceable to a decision, a stage and a client-approved reason.
A prompt costs capacity in every country it runs in, so an unexamined list is expensive as well as noisy.
Record aliases and approved facts at kickoff. A new alias is passed to the extraction model for new answers; for older answers it can only re-attach stored mentions that match the name exactly.
Treat the starter set as version one, and schedule a review of it.

Why does a long prompt list make onboarding harder?

A long list is harder to defend, harder to review and costs more capacity than a short one, even before any result is read. A prompt is a full question DiscoveredBy runs against the AI engines you track; it is not a keyword and not something a user typed (see Key terms).

Each prompt is tracked per location, audience and language, and each combination is its own prompt target that runs separately. Adding a country or reactivating a paused target uses one of your plan's prompt slots, and those slots are pooled across every project the project's owner owns (Plans and limits). If your agency owns several client projects, a client with 40 casually chosen prompts in three countries takes slots that another client might need. If the client owns the project, the slots come from the client's own plan.

There is a second cost that is easy to miss: review. Somebody has to read the results, and a list nobody can explain produces charts nobody can act on. The habit that avoids both is to make the client justify each prompt at kickoff. If you inherit a bloated list, see the companion post Audit your prompt list before adding more prompts.

Flow from a client kickoff interview through four intake blocks to the four places the answers are configured. — The intake feeds four setup areas, in this order.

What should a kickoff intake collect?

A kickoff intake should collect five things: the business goal and scope, the approved facts, the brand's names, the competitors, and the buyer decisions. Each maps to a place in the product, so nothing you ask for is wasted.

Goal and scope. The project has a business type, a description of what the client sells and a plain-English business goal. Several of the product's agents read the goal as steering context rather than as an instruction to follow literally, including the Growth Advisor, the llms.txt Advisor, Optimizations, Articles and the research pass behind prompt suggestions (Projects). It is worth writing it well. Scope also means countries: every country you add is tracked as its own separate set of prompt executions.

Approved facts. These are statements about the client's own brand that the client signs off, such as what a plan costs or which countries it serves. Fact check keeps them as facts in six categories (pricing, company, product, availability, policy, other), each one line of 5 to 300 characters, and checks whether engine answers contradict them: answers to the tracked prompts you select for checking, and answers to its own study questions. It is available depending on the project owner's plan (Fact check).

Brand names and aliases. An alias is another name for a brand. It can be a product name, an abbreviation, a common misspelling or a domain, and every saved alias is passed to the extraction model as another name for that brand (Brands and sub-brands). Clients rarely know how many names they have until you ask.

Competitors. Ask for the competitors the client's sales team actually loses to, not the ones the founder dislikes. Tracked competitors use plan slots too, so a short list is the honest starting point.

Buyer decisions. This is the block that becomes the prompt list, and it deserves the most interview time. The next section shows how.

How do you turn buyer decisions into a small prompt set?

Ask the client for the decisions their buyers make, then write one prompt per decision at each stage where the client can influence the outcome. That gives you a grid, not a brainstorm.

DiscoveredBy lets a prompt carry four optional classifications from closed lists: intent, buyer stage (awareness, consideration or decision), theme (for example competitor alternative, pricing, use case, implementation) and branded or unbranded (Prompts). You can use those same labels as the skeleton of your intake, so the client's answers are already sorted when you enter them.

A workable method:

Ask the client's sales and support leads for the last ten questions prospects asked before buying, and the last ten reasons deals were lost.
Rewrite each as a full question a buyer might put to an AI engine, in the buyer's words.
Label each with a stage and a theme. Merge near duplicates.
Keep the prompts that meet two tests: the client would act on the answer, and you can say which stage it covers.
Mark which prompts need a second country. Most will not.

The result should cover the stages unevenly on purpose. Decision-stage questions (pricing, "alternatives to", "is X good for Y") usually matter most to a sales team, so spend your slots there first. That is a judgment about priorities, not a finding about how engines behave.

Two more inputs help. Once the project exists, a profiler reads the client's site and stages suggested prompts that you accept one at a time; nothing is tracked until you accept it (Onboarding). If Search Console is connected, the client's best-performing queries are handed to the research run that writes those suggestions as extra context, though a query never becomes a suggestion by itself (Suggestions and opportunities). Treat both as material for the interview, not replacements for it. For choosing wording, see How to choose AI tracking prompts that reflect buying decisions.

The Prompts screen with the Import file button and the classification chips shown on tracked prompts. — Demo data. The Prompts screen, where the starter set is entered or imported.

The onboarding questionnaire

Copy this into your own document and send it before the kickoff call. It is written so the client can answer in plain language, and each section says where the answer goes.

CLIENT ONBOARDING: AI VISIBILITY MONITORING
Client:                     Date:            Completed by:

A. GOAL AND SCOPE  (goes to the project record)
A1. Exact website domain to track (cannot be changed later):
A2. What do you sell, and to whom? (one or two sentences):
A3. What is the business trying to achieve in the next 12 months?
    (plain English; this becomes the business goal):
A4. Which countries do buyers ask from? Tick only those that matter:
A5. Which languages must we cover, if not the default?
A6. Which buyer groups (personas) differ enough to track separately?

B. APPROVED FACTS  (goes to the Fact check list)
List the facts that must always be stated correctly (start with the
ten that matter most). One line each,
5 to 300 characters, with a category: pricing / company / product /
availability / policy / other.
B1. Fact:                       Category:        Approved by:
B2. ...
B3. Which facts change often, and who tells us when they change?

C. NAMES  (goes to Settings, Brands)
C1. Company name, exactly as it should appear:
C2. Product names and sub-products:
C3. Abbreviations, old names, and names used by customers:
C4. Common misspellings you have seen:
C5. Other domains you own or have owned:
C6. Names that belong to someone else and could be confused with yours:

D. COMPETITORS  (goes to Competitors)
D1. The three competitors you most often lose deals to:
D2. Names and aliases for each:
D3. Anyone you do NOT want treated as a competitor, and why:

E. BUYER DECISIONS  (goes to Prompts)
E1. Last ten questions prospects asked before buying:
E2. Last five reasons you lost a deal:
E3. Objections you hear most often:
E4. What does a buyer decide first, second and last?
E5. For each question above, which stage is it?  Awareness /
    Consideration / Decision

F. ACCESS AND REVIEW  (goes to Team)
F1. Who should read results only?
F2. Who should be able to edit prompts and competitors?
F3. Who approves facts and receives corrections?
F4. Review date for the starter prompt list:

Two rules make the questionnaire work. First, an answer left blank is a finding: if the client cannot name their abbreviations or approve a pricing fact, you have learned that their own team disagrees, and that is worth knowing before an AI answer reveals it. Second, section F is not paperwork. Membership is per project, and only an owner can add members or change roles, so decide up front who holds that role.

How do you set the project up from the intake?

Enter the intake in a fixed order, so that names and facts exist before the first answers arrive. Set up the project, then the names, then the facts, then the prompts, then access.

Step	Intake block	Where it goes	What to check
1	A	Onboarding, then Settings for the project record	Website is right; countries match A4
2	C, D	Settings, Brands and Competitors	Aliases saved for own brand and each competitor
3	B	Fact check	Each fact is one line and client-approved
4	E	Prompts (add, or import a CSV or Excel file)	Each prompt has a stage and a reason
5	F	Team	Roles match the client's answers

The onboarding flow itself is four steps: website, business, prompts and (optionally) competitors, and it requires at least one prompt before you can continue (Onboarding). Because of that minimum, enter your single highest-priority decision prompt during onboarding, finish steps 2 and 3, then add the rest. Do not treat the minimum as a target: add the smallest defensible set and grow it only after a review.

If your intake is already a spreadsheet, the prompt import takes a file with a column of prompt text and optional country and tags columns. The preview writes nothing and shows what each row would create and how many slots it would use, so you can check the slot cost before you commit. A country in the file must already be enabled for the project (Prompts).

On roles: a viewer can read all of a project's data, an editor can also change it, and only an owner can add members or edit the project record. Exports are gated by the owner's plan, not by role (Team access). For a fuller treatment of access, see Give clients and contractors the access they actually need.

Worked example: Northfold Digital onboards Quillstone

Illustrative example: Quillstone and its competitors are fictional, and the numbers are made up to show the method. Northfold Digital is a fictional agency.

Northfold Digital runs the kickoff with Quillstone, a fictional company that sells document-review software to mid-sized legal and compliance teams. Quillstone's sales lead returns the questionnaire.

Goal and scope. Business goal: "Be shortlisted more often by compliance teams evaluating review software." Countries: the United States and the United Kingdom.

Names. Company: Quillstone. Product: Quillstone Cloud, abbreviated QCloud. A customer misspelling: "Quilstone". Northfold saves the product as a sub-brand (available depending on plan) and the misspelling as an alias. Competitors: Brieflane, Clausewise and Docket North, each with one alias.

Facts. Six candidate facts, but the sales lead cannot approve one of them: whether a free trial exists. Northfold enters the other five and flags the sixth to the client. This is the questionnaire working, not failing.

Decisions. The sales lead supplies 31 candidate questions from calls and lost deals. Northfold rewrites them as buyer questions, merges duplicates, and labels stage and theme. After the two tests, 12 prompts remain:

Stage	Prompts	Example theme
Awareness	3	use case
Consideration	5	comparison, implementation
Decision	4	pricing, competitor alternative
Total	12

All 12 run in the United States. Only the 4 decision-stage prompts also run in the United Kingdom, because that is where the client sells and loses deals. Every prompt uses the General audience and "As written" language, so each country is one slot: 12 + 4 = 16 slots. The other 19 candidate questions go into a parked list with a reason for each, so they can be promoted at the review.

Access. The client's marketing lead is a viewer, Quillstone's content manager is an editor, and Northfold holds the owner role. The review date is set 30 days after the first results.

Northfold now has a prompt list it can defend in one sentence per row: "the sales lead asked this before buying" or "this is a reason a deal was lost."

What this approach cannot tell you

An intake produces a defensible starting point. It does not produce a representative sample of what buyers ask.

Your prompts are a chosen sample. A tracked prompt is a question you selected, run on a schedule. Results describe those questions, not everything buyers type.
The client's memory is biased. Sales teams can over-remember loud deals. Compare the intake to real signals (Search Console queries, support tickets) when you have them.
Aliases help recognition, not certainty. The extraction model decides whether a name in an answer refers to the brand, so a saved alias is a strong hint, not a guarantee.
Approved facts define what is checked, not what is true in the world. Only facts you record are compared, and facts about competitors are out of scope.
Small sets hide gaps. A tight prompt list is easier to explain, but a question you left off is a question nobody is measuring. Keep the parked list, and review it on a date.

Common mistakes:

Starting from a keyword list and calling the terms prompts.
Copying every competitor the client mentions into tracking.
Skipping aliases, then reading a low first-week mention count without knowing names were missed.
Approving facts nobody at the client owns.
Promising outcomes in the kickoff. Nothing here shows that monitoring changes what engines say; scope the work around what you can control, as in Scope an AI visibility engagement around deliverables you can control.

A tidy short stack of index cards on a desk beside a tray holding a few parked cards, one card outlined in coral. — Keep the starter set short and park the rest with a reason.

Frequently asked questions

How many prompts should a new client start with?

There is no correct number. Start with the smallest set where every prompt maps to a decision, a stage and a reason the client cares. Count slots before you commit: each combination of location, audience and language is one slot, and slots are pooled across every project the same owner holds.

Should I use the suggested prompts instead of interviewing the client?

Use them as input, not as the plan. Suggestions come from reading the client's site and, where connected, Search Console, and they only join the tracked set when you accept them. The interview tells you which questions the client would act on.

Do I need to collect brand aliases before the first scan?

Ideally yes. A saved alias is handed to the extraction model for new answers. Adding one later also moves past untracked mentions that match the name exactly, but it does not re-run extraction on old answers, so aliases saved at kickoff give you the cleanest first results.

Who should own the project, the agency or the client?

That is a commercial and account decision, and this post does not settle it. Know the consequences: only an owner can add members, edit the project record or archive it, and plan-gated features follow the owner's plan.

When should we revisit the starter prompt list?

Set a date at kickoff, such as 30 days after first results, and review then. Bring the parked questions and any new sales objections, and open Prompt coverage to see which running prompts you could pause without losing what they regularly see. Change the list deliberately and note what changed, so later comparisons stay readable.

Next step

Once the intake is complete, create the project, save the names and add the first prompts, then check the first results against your questionnaire. Sign in to DiscoveredBy and start with the project record. For the definition, see the prompt glossary entry.

Onboard a client into AI visibility monitoring without a sprawling prompt list