---
title: "White-Label AI SEO Glossary: 19 Terms That Separate a Service From a Dashboard"
description: "A practitioner’s glossary of white-label AI SEO proposals: what tracked prompts, citations, publishing and pricing buy—and what to ask before you resell them."
url: "https://groas.com/post/the-white-label-ai-seo-glossary-22-terms"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/6a8ef129-fea8-4bdb-b7ef-74b3e5b62ae8.png"
published: "2026-10-11T05:20:32.467Z"
modified: "2026-10-11T05:20:32.528Z"
---

October 11, 2026 · 12 min read

# White-Label AI SEO Glossary: 19 Terms That Separate a Service From a Dashboard

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![An open red toolbox holds only flat photo cutouts of tools, suggesting a vendor that sells a picture of the work instead of doing it.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/6a8ef129-fea8-4bdb-b7ef-74b3e5b62ae8.png)

In this article

1. [Monitor vs recommend vs execute](#monitor-vs-recommend-vs-execute)
2. [Tracked prompts (and prompt caps)](#tracked-prompts-and-prompt-caps)
3. [Share of answer / brand mention rate](#share-of-answer--brand-mention-rate)
4. [Verified vs unverified AI crawler hits](#verified-vs-unverified-ai-crawler-hits)
5. [Live-answer fetches](#live-answer-fetches)
6. [Answer mapping](#answer-mapping)
7. [Content gap (for AI citation)](#content-gap-for-ai-citation)
8. [Citation building](#citation-building)
9. [Entity and knowledge graph](#entity-and-knowledge-graph)
10. [Direct-to-CMS publishing](#direct-to-cms-publishing)
11. [Technical fixes for AI visibility](#technical-fixes-for-ai-visibility)
12. [Edge proxy setup](#edge-proxy-setup)
13. [White label vs co-branded vs referral](#white-label-vs-co-branded-vs-referral)
14. [Multi-domain workspace](#multi-domain-workspace)
15. [Per-domain flat rate](#per-domain-flat-rate)
16. [Per-prompt or per-keyword pricing](#per-prompt-or-per-keyword-pricing)
17. [Done-for-you vs done-with-you](#done-for-you-vs-done-with-you)
18. [Overage and hidden seat fees](#overage-and-hidden-seat-fees)
19. [AEO vendor demo questions](#aeo-vendor-demo-questions)
20. [AI visibility score](#ai-visibility-score)

Most agencies won’t tell you this before they resell AI SEO: the proposal sounds like a service, but half its vocabulary describes a screenshot. I’ve sat through those demos. The vendor shows a visibility graph climbing, names 400 tracked prompts, mentions citations and entities, and by slide eight you’re mentally adding a $1,500 retainer. Then you sign, and your team gets a login.

This glossary puts the words back where they belong. I follow the order an agency usually meets them: measurement, work on the site, deployment, then the bill. For each term, ask whether you’re buying someone who watches AI answers or someone who changes what the engines can find. **Only the second belongs in a service retainer without your team doing the work.**

## Monitor vs recommend vs execute

**Monitor records what appears in AI answers; recommend identifies work to do; execute does the work and records what changed.**

I put every white-label proposal into one of those three buckets before I read the price. A monitor reruns questions and graphs the answers. A recommendation tool points out missing pages or citations. An execution service publishes, fixes and builds, then shows an action log. Most proposals I see sell the first bucket with third-bucket verbs: _track, optimize, build_. A missing citation becomes an “action plan.” Nothing got published, fixed or earned. When you resell that as SEO, your client hears action and you bought a screenshot.

![Vendor presents a rising visibility chart while an agency owner asks for an action log](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/cb50009b-aa13-42e8-af52-92ad1d56a432.png)

## Tracked prompts (and prompt caps)

**Tracked prompts are the fixed questions a tool reruns to check whether your client appears in AI answers; a prompt cap limits how many you can check.**

This is where white-label math starts to hurt. Entry tiers often allow 15 to 100 prompts; premium tiers can allow 400, with extra blocks costing more. I’ve seen a $29 plan with 15 prompts and a $489 plan with 400 across four engines, with Claude and Gemini as paid add-ons. A prompt count is a technical limit, not a business result. Add locations, service lines and questions across ten clients, and you can need 2,000 prompts before anyone has changed a page.

**Ask:** What happens at prompt 101? Do old prompts stop running, or does the invoice jump? Are prompts counted per domain or per workspace?

## Share of answer / brand mention rate

**Share of answer is your client’s brand mentions divided by all brand mentions counted across the same prompt set.**

One common calculation is `your brand mentions ÷ all brand mentions × 100`. Presence share is different: it’s the share of tested questions whose answers include your client at all. Those denominators matter. Add weak competitors and a brand’s share can look stronger; test only comparison questions and it can fall. I want the raw counts and a split between branded, comparison and problem questions. If a client appears when asked by name but vanishes in comparisons, an overall percentage won’t tell the team what to fix.

**Ask:** Which questions and competitors went into this number?

## Verified vs unverified AI crawler hits

**A verified AI crawler hit has evidence behind the bot’s identity; an unverified hit is a log entry whose user agent merely claims to be an AI crawler.**

A user-agent string is easy to claim. Verification can involve reverse DNS or CDN security logs, rather than taking the label in an origin log at face value. Access also depends on more than robots.txt: in a [crawler access test](https://dev.to/mahirhir/your-cdn-can-overrule-robotstxt-finding-the-layer-that-refuses-ai-crawlers-34kg), 19 of 79 reachable sites returned a 403 to AI crawler user agents while serving a normal browser a 200; 15 of those 19 allowed the crawler in robots.txt. The refusal appeared to be elsewhere in the delivery stack.

So when a proposal boasts “2,400 AI bot visits last month,” I want to know how many passed its verification method. **Ask for verified and unverified counts separately**, split by engine. Don’t price a service around a user-agent tally.

## Live-answer fetches

**A live-answer fetch is a page request made while an engine is building an answer, rather than a background crawl or a lookup in stored data.**

Vendors blur those events because an index check and a live question are not the same operation. “Coverage across eight engines” may describe different kinds of checks across those engines ([how vendors distinguish coverage from live checks](https://contextbolt.com/blog/ai-visibility-score/)). Index presence can tell you a page is available in stored data. A live fetch gives you evidence it was considered during that answer. Neither, alone, proves it was cited.

**Ask:** Which engines were asked live this week, and which results came from stored data?

## Answer mapping

**Answer mapping assigns each tracked question to the page on the client site that should answer it.**

Good mapping is boring. List 80 prompts, point each at a URL, then mark the thin pages and the questions with no suitable page at all. A keyword sheet renamed an “answer map” doesn’t do that work. If “best CRM for plumbers” points to `/blog/crm-tips` from 2021, the map may be telling you where the problem is, not where the answer lives.

**Ask:** Show me one question, its mapped URL and the edit made after mapping. Without that last step, it’s a plan.

## Content gap (for AI citation)

**An AI citation content gap is a relevant question where competitors appear in tested answers and your client does not, tied to a missing or thin page.**

That isn’t just a keyword gap with _AI_ pasted on it. A keyword report might say you rank 14th. A citation-gap report might show that, across five tests of “best dispatch software for HVAC,” three competitors appeared and your client did not. The distinction matters only if someone turns the finding into a page worth retrieving.

One [review of an AEO tool](https://blog.hubspot.com/marketing/peec-ai-alternatives) describes an Actions tab that surfaces gaps but doesn’t supply briefs, drafts or publishing to close them. That’s useful inventory. It isn’t delivery. **A gap needs an owner, a draft date and a publish log** before I’ll call it progress.

## Citation building

**Citation building means earning relevant third-party pages that AI engines can retrieve and cite when they answer questions about your client’s market.**

Think trade press, review sites, directories with real traffic and vendor comparisons. Don’t confuse those with a bulk submission to 300 directories. A vendor can report “50 citations built” when it means 50 business profiles, none of which appears in a tracked answer. Those listings may serve another purpose, but the proposal shouldn’t sell them as demonstrated AI citations.

**Ask:** Show me three pages you built or earned that appear as sources in live AI answers, alongside the answer text. If the rep shows profile URLs instead, you know what they counted.

## Entity and knowledge graph

**An entity is a distinct company, person or product; a knowledge graph records such things and their relationships.**

For an agency, the practical question is whether a client has a clear, consistent identity across its site and other profiles. `Organization` or `Person` schema and `sameAs` links can help describe that identity ([entity-based SEO practitioner breakdown](https://www.kinexmedia.com/blog/how-does-entity-based-seo-help-brands-rank-in-ai-search/)). They are not a substitute for checking what actually appears in answers. I’ve seen _entity map_ used as a grander name for a keyword list. Names on a slide don’t resolve anything.

**Ask:** Which identity details did you check, what did you change on the site, and how will you inspect the result?

## Direct-to-CMS publishing

**Direct-to-CMS publishing means the vendor can stage and publish or update pages inside the client’s content system without handing your team a ticket.**

This is the cleanest boundary between recommend and execute. A dashboard that exports a brief still needs your copywriter, your QA process and someone with publishing access. An executed edit lands on a URL, and an action log records what changed and why. “CMS integration” can also mean exporting a Google Doc or CSV. I ask who holds access and who clicks publish. If the answer is your team, put your labor in the retainer calculation, not theirs.

## Technical fixes for AI visibility

**Technical fixes remove obstacles that keep a page from being fetched, parsed or used: blocked crawlers, bad status codes, messy HTML, weak internal links or missing relevant markup.**

The work should be specific. Check robots.txt, then check whether the CDN still returns a 403. Inspect the template that renders an answer as JavaScript soup. Fix relevant schema and links where they’re wrong. What the deck calls _AI optimization_, I call server logs plus HTML hygiene. A generic audit renamed an “AI readiness score” isn’t the same as a fix.

**Ask:** Which five URLs will you address first, what will change, and how will you check that a fetch now works?

## Edge proxy setup

**An edge proxy is a layer in front of the client site that can check crawler identity, handle requests before they reach the origin and log what was fetched.**

Think bouncer with a guest list. Properly configured, it can help distinguish claimed bots from verified ones and expose blocking that an origin log won’t explain. But _edge proxy_ is a deployment term, not proof that citations improved. I want to see what it was configured to do, what the logs show by engine and what happens when a legitimate fetch fails.

![Cutaway diagram of an edge proxy checking AI crawler requests before they reach a website](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/0559fe92-bc3b-4b79-a9b0-f2c724ad3812.png)

## White label vs co-branded vs referral

**These are three ways to sell another provider’s work, distinguished by whose brand and client relationship stay visible.**

- **Referral:** You send the client to the vendor for a fee; the vendor owns the contract and data.
- **Co-branded:** Both names remain visible.
- **White label:** Your brand, reports and support face the client while the provider stays behind the scenes ([white-label packaging explained](https://thebestreputation.com/online-reputation-management/white-label-reputation-management/)).

A logo swap isn’t full white label if the client still enters a vendor domain, receives vendor emails and exports vendor-branded PDFs. I’ve seen that movie. The client googles the footer and your markup evaporates. **Ask: whose domain, whose email, whose invoice?**

## Multi-domain workspace

**A multi-domain workspace lets an agency manage client domains in one place while separating their prompts, reports and permissions.**

It’s what turns a single-site tool into something an agency can operate. Agency packaging can include multiple workspaces, prospect workspaces and one invoice across clients ([an example pricing page](https://otterly.ai/pricing)). Watch for vendors selling seats as if they were workspaces: five logins looking at the same 100 prompts don’t give you five separate client setups. Fine for one in-house team. Awkward for twelve retainers.

**Ask:** Can I isolate prompts and reports per domain, and can I open a prospect workspace without another subscription?

## Per-domain flat rate

**A per-domain flat rate is a fixed monthly base price for each client domain, subject to whatever limits the contract puts around it.**

Say the domain costs $199 a month and you resell the work at $800. That base makes a retainer easier to plan than a bill that rises with every new question. But _flat_ can have footnotes: publishing volume, citations or engine coverage may still be metered. I price it the way I used to price PPC management: flat fee, named deliverables, overage in writing.

**Ask:** What’s included in that fee, in numbers, and what triggers the next dollar?

## Per-prompt or per-keyword pricing

**Per-prompt or per-keyword pricing charges for questions or keywords checked, not for changes made to the site.**

You’ll see this model across vendors such as Peec, Profound, Scrunch and Otterly. Its risk for a retainer is plain: [prompt volume is a technical constraint rather than a business metric](https://blog.hubspot.com/marketing/peec-ai-alternatives). A client starts with 50 questions; then you add locations and services and need 200. If the charge scales with prompt count, your cost rises while your agreed retainer doesn’t. I’ll buy that tracking for diagnostics. I won’t bury it in a fixed-price service without overage math.

**Ask:** What do 100 extra prompts cost, and do old ones pause or does the bill rise automatically?

## Done-for-you vs done-with-you

**Done-for-you means the vendor completes the publishing, fixes and citation work; done-with-you means your team receives recommendations and completes it.**

Proposals blur this with verbs like _optimize_, _enhance_ and _activate_. I use a calendar test. Done-for-you work has shipped URLs and dated actions inside the fee. Done-with-you work has a prioritized ticket list. Both can be useful, but only one removes the labor from your agency’s retainer. As [agency AEO pricing shifts from reporting toward execution](https://groas.com/post/from-white-label-seo-to-white-label-aeo), that distinction decides what you can promise without adding headcount.

**Ask:** Show me last week’s log for one domain: URLs changed, not tasks suggested.

## Overage and hidden seat fees

**Overages and seat fees are charges beyond the headline price for more prompts, engines, users or domains.**

A $29 starter can become a much larger invoice once engines are add-ons and the 16th prompt pushes you into another tier. Seats catch agencies too: a strategist, a client and a view-only founder may each need access. I ask for the worst-case invoice for the book I expect to manage, not the prettiest number on the pricing page.

**Ask:** What do ten domains, three seats and 150 prompts each cost with every engine you need? Which limit trips first, and what’s the unit price after it?

## AEO vendor demo questions

**AEO vendor demo questions are the short checks that reveal whether a proposal sells monitoring, recommendations or executed work.**

I use these five before signing, alongside the [buying questions I use for AEO vendors](https://groas.com/post/aeo-for-agencies-the-questions-i-keep-ge) and the [practical questions on pricing, proxy setup and publishing](https://groas.com/post/agency-owners-asked-chatgpt-these-8-ques):

1. Show me last week’s action log for one domain, with URLs changed.
2. Which prompts were asked live, and which were scored from stored data?
3. How many crawler hits were verified, split by engine?
4. What do 150 prompts, three seats and ten domains cost in one number?
5. Whose domain, email and invoice does the client see?

**Start with the log.** It makes the other four answers easier to interpret.

## AI visibility score

**An AI visibility score summarizes how often a client appears across a vendor’s chosen test questions and engines.**

That’s why the impressive graph from the opening needs a closer look. Vendors choose different questions, engines and scoring methods, so a 31 in one tool and a 68 in another don’t tell you which service did better. A rising score tells you what happened in that tool’s tests. It doesn’t tell you who changed a page, whether an answer cited it or whether your client earned revenue. I used to report scores like this for rankings. I was wrong to let clients read them as outcomes.

**Ask:** Show me the questions, the live checks and the dates. Then show me the action log. If I could retire one term tomorrow, it would be _AI visibility score_. Not because measurement is useless, but because one number makes watching look like fixing. A page published, a fetch verified, a citation earned inside a live answer: those leave traces. Ask for those first, and the glossary gets very short, very fast.

## Frequently Asked Questions

### What's the difference between monitoring AI answers and actually doing SEO work?

Monitoring records what appears in AI answers, recommending identifies work to do, and executing does the work and records what changed. Many proposals sell monitoring with execution verbs like track and optimize, so the client hears action while nothing was published, fixed or earned.

### How many tracked prompts does an AI SEO tool actually need?

A prompt count is a technical limit, not a business result. Entry tiers often allow 15 to 100 prompts and premium tiers around 400, but adding locations, service lines and questions across ten clients can require 2,000 prompts before anyone has changed a page.

### What does share of answer mean in AI search tracking?

Share of answer is your client's brand mentions divided by all brand mentions counted across the same prompt set, usually multiplied by 100. The denominator matters: adding weak competitors can inflate it, and testing only comparison questions can lower it, so ask which questions and competitors went into the number.

### How can you tell if an AI bot visit is real?

A verified AI crawler hit has evidence behind the bot's identity, such as reverse DNS or CDN security logs; an unverified hit is a log entry whose user agent merely claims to be an AI crawler. A user-agent string is easy to claim, so ask for verified and unverified counts separately, split by engine.

### What is a content gap for AI citations?

An AI citation content gap is a relevant question where competitors appear in tested answers and your client does not, tied to a missing or thin page. It differs from a keyword ranking report, and it only matters if someone turns the finding into a page worth retrieving, with an owner, a draft date and a publish log.

### What does citation building actually mean for AI SEO?

Citation building means earning relevant third-party pages that AI engines can retrieve and cite, such as trade press, review sites, directories with real traffic and vendor comparisons. A vendor can report 50 citations built when it means 50 business profiles that never appear in a tracked answer, so ask for pages that appear as sources in live AI answers.

### What does direct-to-CMS publishing mean in a white-label proposal?

It means the vendor can stage and publish or update pages inside the client's content system without handing the agency a ticket. CMS integration can also just mean exporting a Google Doc or CSV, so ask who holds access and who clicks publish; if it's the agency's team, that labor belongs in the retainer calculation.

### Is an AI visibility score a reliable measure of results?

An AI visibility score only summarizes how often a client appears across a vendor's chosen test questions and engines. Different tools pick different questions and scoring methods, so a 31 in one tool and a 68 in another don't show which service did better. Ask for the questions, the live checks, the dates and the action log instead.

## Pay For Results, Not For Hours

Businesses buy the outcome, agencies resell it, and groas answers for it either way.

[See If You Qualify](https://groas.typeform.com/to/xC1bQNUT)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms#webpage","url":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms","name":"White-Label AI SEO Glossary: 19 Terms That Separate a Service From a Dashboard","description":"A practitioner’s glossary of white-label AI SEO proposals: what tracked prompts, citations, publishing and pricing buy—and what to ask before you resell them.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/6a8ef129-fea8-4bdb-b7ef-74b3e5b62ae8.png"},"breadcrumb":{"@id":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms#breadcrumb"},"dateModified":"2026-10-11T05:20:32.528Z","mainEntity":{"@id":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"White-Label AI SEO Glossary: 19 Terms That Separate a Service From a Dashboard","item":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms#article","url":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms","mainEntityOfPage":{"@id":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"White-Label AI SEO Glossary: 19 Terms That Separate a Service From a Dashboard","description":"A practitioner’s glossary of white-label AI SEO proposals: what tracked prompts, citations, publishing and pricing buy—and what to ask before you resell them.","datePublished":"2026-10-11T05:20:32.467Z","dateModified":"2026-10-11T05:20:32.528Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/6a8ef129-fea8-4bdb-b7ef-74b3e5b62ae8.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"client, prompts, ask, answer, white-label, seo, service, glossary","wordCount":2495,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"Ex Goldman Sachs and Ex Stanford Computer Science","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/the-white-label-ai-seo-glossary-22-terms#faq","mainEntity":[{"@type":"Question","name":"What's the difference between monitoring AI answers and actually doing SEO work?","acceptedAnswer":{"@type":"Answer","text":"Monitoring records what appears in AI answers, recommending identifies work to do, and executing does the work and records what changed. Many proposals sell monitoring with execution verbs like track and optimize, so the client hears action while nothing was published, fixed or earned."}},{"@type":"Question","name":"How many tracked prompts does an AI SEO tool actually need?","acceptedAnswer":{"@type":"Answer","text":"A prompt count is a technical limit, not a business result. Entry tiers often allow 15 to 100 prompts and premium tiers around 400, but adding locations, service lines and questions across ten clients can require 2,000 prompts before anyone has changed a page."}},{"@type":"Question","name":"What does share of answer mean in AI search tracking?","acceptedAnswer":{"@type":"Answer","text":"Share of answer is your client's brand mentions divided by all brand mentions counted across the same prompt set, usually multiplied by 100. The denominator matters: adding weak competitors can inflate it, and testing only comparison questions can lower it, so ask which questions and competitors went into the number."}},{"@type":"Question","name":"How can you tell if an AI bot visit is real?","acceptedAnswer":{"@type":"Answer","text":"A verified AI crawler hit has evidence behind the bot's identity, such as reverse DNS or CDN security logs; an unverified hit is a log entry whose user agent merely claims to be an AI crawler. A user-agent string is easy to claim, so ask for verified and unverified counts separately, split by engine."}},{"@type":"Question","name":"What is a content gap for AI citations?","acceptedAnswer":{"@type":"Answer","text":"An AI citation content gap is a relevant question where competitors appear in tested answers and your client does not, tied to a missing or thin page. It differs from a keyword ranking report, and it only matters if someone turns the finding into a page worth retrieving, with an owner, a draft date and a publish log."}},{"@type":"Question","name":"What does citation building actually mean for AI SEO?","acceptedAnswer":{"@type":"Answer","text":"Citation building means earning relevant third-party pages that AI engines can retrieve and cite, such as trade press, review sites, directories with real traffic and vendor comparisons. A vendor can report 50 citations built when it means 50 business profiles that never appear in a tracked answer, so ask for pages that appear as sources in live AI answers."}},{"@type":"Question","name":"What does direct-to-CMS publishing mean in a white-label proposal?","acceptedAnswer":{"@type":"Answer","text":"It means the vendor can stage and publish or update pages inside the client's content system without handing the agency a ticket. CMS integration can also just mean exporting a Google Doc or CSV, so ask who holds access and who clicks publish; if it's the agency's team, that labor belongs in the retainer calculation."}},{"@type":"Question","name":"Is an AI visibility score a reliable measure of results?","acceptedAnswer":{"@type":"Answer","text":"An AI visibility score only summarizes how often a client appears across a vendor's chosen test questions and engines. Different tools pick different questions and scoring methods, so a 31 in one tool and a 68 in another don't show which service did better. Ask for the questions, the live checks, the dates and the action log instead."}}]}]}
```