Compare the unit before comparing the price
An AI visibility monitoring quote is useful only when its units match the work you need. A tracked prompt can be a continuing allocation, a country-specific configuration, or an input to a model-by-model credit calculation. None of those automatically equals one usable answer. Before comparing subscriptions, write down the questions, markets, collection surfaces, frequency, evidence requirements and access rights. Convert that design into scheduled observations, then ask each supplier to map its billing rules to the same design. Keep commercial capacity, collection attempts, complete answers and comparable answers in separate columns. This makes the quote reviewable without pretending that every captured response has the same value.
The quote that looks smaller may cover a different job
An export marketing team wants to know how buyers discover suppliers in two markets. It receives one quote for a number of prompts and another for a number of credits. A third proposal combines monitoring with monthly content work.
Dividing each subscription price by the largest number on its sales page produces an attractive spreadsheet. It does not produce a fair comparison. One number may describe how many questions can remain active. Another may describe an ongoing allocation across several models. The third may include editorial work that the first two do not offer in that package.
The purchasing decision has two parts: whether the proposed observation design answers the team's questions, and what the supplier charges to operate that design. Combining those questions too early hides exclusions. A low unit price cannot repair collection from the wrong surface or missing evidence.
Seven units that belong on different lines
Generative Engine Optimization, or GEO, concerns how a business is represented in generated answers and AI-assisted discovery. Monitoring that representation requires more detail than a keyword count.
| Unit | Working definition | What to clarify |
|---|---|---|
| Question text | The exact wording submitted to an AI surface | Does an edit create a new version? |
| Monitoring configuration | A question combined with the relevant surface, market, language and mode | Which settings create another billable configuration? |
| Active prompt slot | Capacity reserved for an active question under a supplier's rules | Is the country part of the slot? |
| Credit allocation | Capacity expressed in a supplier-defined unit | Does it remain allocated or get consumed? |
| Scheduled observation | One planned collection for one configuration at one time | Are all dates and surfaces included? |
| Collection attempt | An actual attempt, including a failed or repeated attempt | Do retries incur charges? |
| Comparable answer | A complete retained answer that meets the study's comparison rules | Can its surface and environment be checked? |
These definitions are a procurement worksheet, not an industry standard. A supplier may use the words differently. The contract and a worked example must settle the meaning.
A complete answer can still be unsuitable for a particular comparison. For example, an API response and a consumer web-search response may both be valid records while representing different experiences. Retain both under their own labels. A financial worksheet should not erase that distinction.
Two official examples show why “prompt” is not enough
OtterlyAI's help page describes an active prompt allowance tied to country settings. The same question configured for the United States and the United Kingdom uses two prompt slots. It also says that connecting a prompt to multiple brand reports does not multiply that prompt's slot count. This describes a capacity rule; it does not establish the price or suitability of every engine configuration. OtterlyAI prompt-counting documentation
Peec AI's agency pricing page describes credits as allocations that stay assigned until the configuration changes. Its example calculates one question on one model across a 30-day month as 30 credits. The page explicitly distinguishes these allocations from a balance that resets each month. These are documented commercial definitions, not measurements of completed answers. Peec AI agency pricing
| Documented rule | Procurement implication | Boundary |
|---|---|---|
| OtterlyAI counts country-specific prompt slots | Map each required question-country combination | Do not infer that all desired surfaces are included |
| A prompt can feed multiple OtterlyAI brand reports without extra prompt slots | Separate report views from unique collection configurations | Confirm client permissions and the actual project setup |
| Peec AI agency credits remain allocated to configured capacity | Model the ongoing configuration and available allocation | Do not treat the displayed allocation as a count of successful responses |
Both examples were checked on October 28, 2026. They are deliberately narrow. We did not log into either product or test billing. A current written quote should confirm the selected plan, engines, countries, frequency and exceptions. The table makes no claim about which product is cheaper or better.
Build one observation schedule before requesting quotes
Suppose a B2B team proposes 20 exact buyer questions, two markets and three named AI surfaces, collected once per day for 30 days. This is a hypothetical planning exercise, not a customer deployment or a Semvi subscription allowance.
The planned observation count is:
20 questions × 2 markets × 3 surfaces × 30 dates = 3,600 scheduled observations.
That formula assumes that every combination is required and supported. It describes physical coverage, not any supplier's invoice. If a vendor represents multiple markets differently, obtain its mapping rather than changing the worksheet to resemble its sales page.
| Schedule variant | Calculation | Planned observations |
|---|---|---|
| All 20 questions collected daily | 20 × 2 × 3 × 30 | 3,600 |
| All 20 questions collected on four specified dates | 20 × 2 × 3 × 4 | 480 |
| Ten core questions daily | 10 × 2 × 3 × 30 | 1,800 |
| Ten exploratory questions on four specified dates | 10 × 2 × 3 × 4 | 240 |
| Combined core and exploratory schedule | 1,800 + 240 | 2,040 |
The combined design is an editorial option for deciding what to observe frequently. It is not equivalent to the daily design and does not imply a particular billing discount. Four observations across a month provide less information about daily variation than 30. A product's weekly billing rule may also differ from the ratio of collection dates.
Write the actual dates into the scope. “Monthly” can mean one report, a calendar billing period, or repeated collection throughout that period. “Weekly” also needs dates; do not silently assume that every month has exactly four collection days.
Separate intended coverage from what actually arrived
At the end of the hypothetical 3,600-observation schedule, imagine the following final states. These numbers are invented solely to demonstrate reconciliation.
| Final state of a scheduled observation | Illustrative count | Treatment |
|---|---|---|
| Complete consumer AI answer retained | 3,200 | Eligible for the applicable answer review |
| Requested AI feature did not produce an answer | 160 | Report separately from answered questions |
| Access or rate limit prevented collection | 100 | Record the access failure |
| Timeout left no complete answer | 80 | Record the incomplete attempt |
| Capture could not be validated | 60 | Exclude from usable-answer totals |
| Total scheduled observations | 3,600 | All scheduled cells accounted for |
These are final states for scheduled cells, not a count of network requests. A retry can produce another attempt for the same cell. Preserve the attempt history, then report one reconciled final state according to the predeclared method. Ask the supplier how retries affect both billing and its dashboard totals.
If only 2,700 of the 3,200 retained answers satisfy the intended comparison rules, the useful denominator for that comparison is 2,700. The remaining 500 answers may remain useful in another labelled group. They should not be deleted to improve a rate or pooled into a comparison whose conditions they do not satisfy.
For a monitoring cost of C, cost per complete retained answer would be C / 3,200, while cost per comparable answer would be C / 2,700 in this invented example. Neither figure establishes return on investment. Recommendation quality and business outcomes require additional evidence.
Geographic coverage needs a collection specification
A question about buying in the United States does not establish a US collection environment. Nor does an English interface establish a particular country. Specify what the supplier means by market coverage: a configured country, visible platform location, verified network exit, language, account context, or a combination.
Some of those conditions may be unavailable for a given surface. An honest quote can mark them unknown and explain the practical limitation. A table with an unexplained country flag cannot resolve it.
For each named surface, request the collection method, search or browsing mode, visible model information, authentication state and the evidence kept about location. Keep consumer ChatGPT, Gemini App, Google AI Overview, Google AI Mode and API calls as distinct configurations when they represent different observation tasks. Coverage of one does not establish coverage of the others.
Shared capacity does not settle client ownership
An agency may monitor several brands against one general question. A single retained answer could support several analyses if the software and client agreements permit that use. Conversely, a question containing client-specific information may require a separate project even when its wording resembles another client's question.
Ask separately about prompt capacity, project limits, client logins, report viewers, export permissions and administrative ownership. A plan with spare prompt capacity may still lack the access arrangement your delivery process needs.
Also check whether a pitch project is temporary, whether historical answers remain accessible after a conversion, and whether moving a client requires an export or a new account. Do not infer those terms from a vendor's ability to display multiple brands. Our guide to monitoring several client brands covers the operational handoff in more detail.
Run a quote-normalization exercise in six steps
- Freeze a small question set. Record exact text, intended buying decision, language and version. Keep questions used to measure natural discovery free of the target brand name.
- List the collection configurations. Name each surface and required environment. Mark unsupported or unverified combinations explicitly.
- Specify the schedule. Use dates, not just daily, weekly or monthly labels. State whether extra checks are optional, scheduled or event-triggered.
- Ask for a commercial mapping. Request the prompt slots, credit allocations, project capacity, seats, surface add-ons, setup charges and implementation work needed for that exact schedule.
- Reconcile a limited pilot. Check scheduled cells against attempts, retained answers, failures and the comparison subset. Inspect a complete export rather than relying on a headline dashboard total.
- Confirm the exit conditions. Establish who can download the answers, citation URLs and metadata; what retention covers; and what access remains when the commercial relationship ends.
A completed worksheet should make two proposals comparable even when their published units differ. It should also expose where they remain incomparable. Leaving an unresolved field visible is better than quietly filling it with an assumption.
Applying this worksheet to Semvi by Xindar
Xindar publishes this article and provides monitoring and export marketing services. The product team confirms that Semvi supports complete-answer and citation-URL export, ChatGPT and Gemini consumer-web sampling, and retention for at least one year, extended to match the agreed service term. Those are product-team statements; this article does not independently test the account workflow.
For a purchase discussion, request the mapping from your proposed schedule to the actual quote. Confirm supported modes, countries, frequency, export fields, overages and the service-end rules. Google Search, AI Overview and AI Mode coverage need separate confirmation. No public prompt quota, credit conversion, retry policy or standalone subscription term is established by the arithmetic in this article.
The Semvi product overview and our screen-to-file export acceptance test provide a starting point for that review. Monitoring and implementation deliverables should have separate line items so the buyer can assess what each part contributes.
Questions procurement teams ask
Does a 100-prompt plan produce 100 answers?
The label alone cannot answer that. Ask how active questions, models, markets and collection dates interact. Then distinguish the planned answer count from completed, retained and comparable answers.
Can I compare two products by cost per answer?
Only after defining the same answer unit and coverage boundary. A consumer answer with retained evidence and an API response collected under different conditions should not become interchangeable because both are called answers. State which costs the calculation includes.
Are weekly checks simply daily checks at a lower price?
They provide a different observation schedule. The price relationship depends on the supplier's commercial rules. Fewer dates also reduce what you can observe about variation between those dates.
Should every failed collection count as no recommendation?
No. A missing or incomplete answer cannot establish whether a brand would have been recommended. Retain the failure and report it alongside the valid-answer denominator. A complete answer that does not recommend the brand remains a valid negative observation.
Does more monitoring make a brand more visible?
Collection creates evidence about sampled answers. It does not by itself change those answers or establish demand. Use the evidence to choose specific website, content, product-information or external-source work, then evaluate later observations without claiming that volume alone caused improvement.
Method and revision note
The original general budget framework remains available in The Next Dollar in GEO: Choosing the Bottleneck to Fix. This later URL originally repeated that article and was substantially revised on October 28, 2026 to answer the narrower monitoring-unit procurement question. Its existing URL is retained.
Official supplier documentation was checked on October 28, 2026. Its definitions support only the statements linked above; they are not independent delivery or billing tests. All schedule, outcome and cost calculations in this article are explicitly hypothetical. They do not report customer performance, Xindar results, current vendor prices or an improvement in natural AI recommendations.