Allocate a limited prompt budget across products and markets

To allocate a limited prompt budget, count the cost of every prompt in slots (locations times audiences times languages), rank your candidate prompts by the buying decision they cover and their cost, and fund the ranked list until you reach your plan's limit, keeping a small reserve. A prompt tracked in three countries for two audiences costs six slots, not one, so the cheapest way to lose coverage is to add variants without a ledger. The worksheet below is a planning method, not a product feature. Its output is a hypothesis about where to look, which you then check against real answers.

In short

• A prompt is not the unit your plan meters. A tracked variant is: one location, one audience and one language. Multiply before you commit.
• Fund by buying decision first, then by market, then by variant. Depth on the decisions that matter usually tells you more than thin coverage of everything.
• Keep a reserve. Accepting a suggested prompt or resuming a paused one needs free slots.
• After two weeks of answers, use the coverage view to find prompts you can pause, and spend the freed slots deliberately.
• A limited budget gives you a sample of your market, not the whole of it. Say so in every report.

What does one prompt actually cost?

A prompt costs one slot for every variant you track, and a variant is one combination of location, audience and language. In DiscoveredBy a prompt is a full question you chose to monitor because your buyers ask it; it is not a keyword (see key terms). A prompt slot is the unit your plan meters. Each variant, which the product stores as a prompt target, uses one.

The arithmetic is multiplicative. A prompt tracked in 3 countries as General plus 2 personas, all in the language "As written", produces 3 x 3 x 1 = 9 targets, according to the prompts documentation. A city is a location of its own, so it takes a slot for each audience and language it runs as. A single prompt tops out at 100 active variants, and every path that adds a variant enforces that cap (languages and templates).

Two more facts shape the plan:

• Slots pool across your account. They are counted once for every project you own together, not reset per project. If you set up one project per product, all of them draw on the same pool (plans and limits).
• Engine choice is free. Restricting a prompt to a subset of engines costs no slots. It only changes how many answers you receive.

Plan figures change, so this post uses none. Read your current slot limit from your billing screen, which shows slot usage against your total, and from the pricing page. See plans and limits for what is metered.

Enlarge image

Which prompts deserve the budget first?

Fund prompts in order of the buying decision they cover, because a decision-stage question sits closest to a purchase. A prompt can carry a buyer stage of awareness, consideration or decision, and an intent such as comparison or commercial (prompts documentation). Those labels give you a ready-made way to sort, but they are optional: a prompt you imported has none until you set it, so classify the candidates first.

A workable order looks like this:

1. Decision-stage and comparison prompts for your main product in your main market. These are the questions where a shortlist forms. For help writing them, see How to choose AI tracking prompts that reflect buying decisions.
2. The same for your second product, so you can compare products fairly.
3. Consideration prompts, one or two per product, to see how the category is described before a shortlist exists.
4. Market variants of your best prompts, added one market at a time.
5. Persona and language variants, only for prompts where the audience or language changes what you would do.

Awareness prompts come last because they are the least specific about your product. That is a judgement, not a rule; a team creating a new category may reverse it.

How do you weigh products against markets?

Weigh a product or market by what a change in visibility would let you decide, not by how much you would like to be seen there. A market where you sell, have local facts and can act on a finding earns slots. A market you might enter someday earns one or two exploratory prompts at most.

Ask three questions of every product and market pair:

• Can you act on a finding? If you cannot change the page, the listing or the outreach in that market, a slot there produces a chart, not a decision.
• Is it a different question? If the prompt reads the same in the UK and the US, and the answers are likely to be similar, a second country adds cost before it adds information. Test with one prompt first.
• Is it comparable? Adding a market is only useful if you can compare it. See Local AI visibility and Track AI visibility across languages for keeping results comparable.

Which variants cost slots without adding answers?

Some variants use a slot and still never run on certain engines, so check engine coverage before you multiply. Per the languages and templates page, the two Google engines skip persona variants, Chinese variants, and prompts over 700 characters once encoded. ChatGPT (app) skips persona variants, city variants, Chinese variants, and prompts over 2,000 characters. Gemini (app) skips persona variants, Chinese variants, and prompts over 2,000 characters, but runs city variants.

The practical consequence: a persona variant is a slot that produces no answers from the two Google engines, ChatGPT (app) or Gemini (app); it runs only on the chat engines. The slot count is by variant, not by engine, so you pay the same for it. If the app and Google engines matter to your buyers, a persona variant buys less than it appears to. Check engines for the current collection details before you decide, and keep prompts short if you want them to run everywhere.

Reserve persona variants for prompts where the audience genuinely changes the question, as covered in Measure AI visibility for different buyer personas.

The slot ledger: a copyable worksheet

Build a ledger with one row per prompt and a cost you can total. The template below assumes your budget is the number on your billing screen. Put it in a spreadsheet; the columns are the ones that matter.

PROMPT BUDGET LEDGER

Budget (slots on my plan, whole account): ____
Reserve I will hold back (accepted suggestions, resumed prompts, new products): ____
Slots I can allocate = budget minus reserve: ____

Columns, one row per prompt:
  ID | Product | Prompt (full question) | Buyer stage | Intent
  | Locations (n) | Audiences (n) | Languages (n)
  | Slots = locations x audiences x languages
  | Value 1-3 (what decision does it inform?)
  | Actionable 1-3 (can we change something if it moves?)
  | Overlap factor (1 = unique, 0.5 = another prompt covers the same ground)
  | Score = (Value + Actionable) x Overlap factor / Slots
  | Funded? (yes / no / later)

Rules for filling it in:
  1. Sort by Score, highest first.
  2. Fund down the list until the running slot total reaches
     the allocatable number. Do not fund partway into a prompt's variants.
  3. Every product and every market you report on gets at least one
     funded prompt, or is written off as "not measured this quarter".
  4. Note every "no" and "later" with a reason. They are the measurement gaps.
  5. Re-run the ledger at each review, not once.

Footer for every report that uses this set:
  "Measured on __ prompts across __ markets and __ products.
   This is a sample, not the whole market."

Enlarge image

The score is a planning heuristic invented for this worksheet. Nothing in DiscoveredBy computes it, and the weights are yours to change. What matters is that the same formula is applied to every candidate, so trade-offs become visible instead of political.

A worked example

Quillstone sells two products: Quillstone Review (document review) and Quillstone Audit (compliance evidence). It sells mainly in the US and UK, and has just opened a small pipeline in Germany. Its account allows 50 slots. It holds 4 in reserve, leaving 46 to allocate. Competitors named in its answers include Brieflane and Clausewise.

The team lists candidate prompts and costs each one in slots:

The arithmetic checks: 12 + 10 + 6 + 4 + 6 + 8 = 46, which equals the allocatable 46, and 46 + 4 in reserve = 50.

Three decisions came out of the exercise:

• The persona block costs 6 slots for 3 prompts, double what the same prompts cost as General only. The team noted that persona variants do not run on the Google engines, ChatGPT (app) or Gemini (app), so it treated the block as a targeted test, not a permanent commitment.
• Germany costs 8 slots for 4 prompts because each runs in two languages. The team kept it because the pipeline was live, but wrote in the report: "Germany measured on 4 prompts; treat as exploratory."
• The team also wanted the four Audit consideration prompts in the UK, which would cost 4 more slots and take the total to 50, over the allocatable 46. That block ranked below the funded blocks, so it was marked "later", with the reason recorded.

Nothing in this example says the allocation is right. It says the trade-offs were written down.

How do you know which prompts to pause later?

After the prompts have run for a while, use the coverage view to find pauses that lose the least, then spend the freed slots on the "later" rows. Prompt coverage compares the last 28 days of answers for the prompts running now and recommends a set you could pause together while every source and brand those prompts see regularly is still seen by a prompt you keep.

The mechanics matter for budget planning:

• A prompt is judged only once it has analysed answers on at least 14 different days in the window. A prompt you added last week will read "not enough answers yet" and is never recommended, so coverage cannot prune a brand-new set.
• Some prompts are never recommended: prompts chosen for fact checks, prompts named by an optimization that is still open or whose outcome is not settled, prompts linked to a citation gap that is accepted, in progress, implemented or monitoring, and the last running prompt for a country, city, persona, language or engine. That last rule works as a minimum-coverage floor for each market you keep.
• You can preview any selection before pausing. The preview shows slots freed, answers removed, and sources or brands that would no longer be seen.
• Pausing keeps history. A paused prompt's variants stop counting against your slots, and resuming needs free slots, so it can be refused if the plan has too few.

Read the limits too. The bars for "regular" (10% of a prompt's answers) and "seen" (2.5%) are fixed. Items are pooled across engines, locations and personas, so a source that only one country's prompt sees can still be judged covered by another country's prompt. Coverage also compares prompts by what their answers contain, not by wording. Treat a recommendation as a starting point for review, not an instruction.

Enlarge image

What if your plan or team changes?

The plans and limits page describes what happens when a trial ends: saved data remains, but only the oldest active projects and prompt slots within your current plan are monitored, and you pause prompts to choose which others run (product trial). That page describes it for an expired trial, so check it before assuming the same for any other change. Either way, re-run the ledger before a limit drops so the oldest prompts are not the ones that matter least.

Editors on a project someone else owns spend that owner's slots. If several people add prompts, agree who owns the ledger. Audit your prompt list before adding more prompts is a useful review to run before you spend reserve.

Common mistakes and what this cannot tell you

• Counting prompts, not slots. Nine slots on one prompt is not the same as nine prompts.
• Adding a market before proving the question differs. Start with one prompt in the new market, compare, then expand.
• Spending the reserve. A budget spent to zero cannot absorb an accepted suggestion or a paused prompt you want to resume.
• Treating the score as a measurement. It ranks your intentions. It says nothing about how AI engines rank anyone.
• Reading a small set as the market. A limited set is a sample. Say how many prompts stood behind every figure.
• Pausing on coverage alone. Coverage judges overlap in answers over 28 days. It does not know which prompts your leadership cares about, and it cannot see future answers.
• Ignoring what a pause does to your trends. A paused prompt stops producing new answers and its earlier answers drop out of the prompt coverage figures, so a trend can shift because the mix of prompts changed, not because of anything an engine did.

Frequently asked questions

How many prompts should I track per product?

There is no correct number, and the documentation does not give one. Start with the decision and comparison questions a real buyer asks, then stop when adding another prompt would repeat a question you already track. Your plan's slot limit sets the ceiling; your ledger sets where you stop.

Does adding a country double my cost?

Going from one country to two doubles that prompt's slots, all else equal, and each further country adds another full set of its audience and language variants. Add countries to your best prompts first and test one before applying it to the whole set.

Do languages and personas draw on a separate allowance?

No. Languages, personas, engine choice and saved templates draw on the same prompt-slot pool as countries, on every plan, according to the languages and templates page. Engine choice itself costs no slots.

Can I import a large list and sort it out later?

You can import from CSV or Excel, but imports add country-wide targets only, and every new country uses a slot. The preview states how many slots the import would consume and how many you have left, and an import that does not fit is refused whole. Import a list you have already costed, not a raw export of everything you could ask. See prompts for the import rules.

Should each product have its own project?

The docs describe a project as one site with its own prompts, competitors and collected data, while prompt slots pool across all projects you own. Splitting into projects therefore does not give you more slots, and the number of active projects is itself limited by your plan. Separate projects make sense when products live on separate sites, not as a way to organise a budget.

How often should I redo the allocation?

At each planned review, and whenever you launch a product, enter a market or change plan. A quarterly pass is a reasonable default; there is no setting that does it for you.

Next step

Open your prompt list, cost every prompt with the formula above, and total it against your plan's limit before you add anything else. In DiscoveredBy, the Prompts list shows each prompt's locations, audiences and languages, the billing screen shows slots used against your total, and Prompt coverage shows slots in use and which pauses lose the least. Start or sign in at DiscoveredBy to cost the prompts you already run.