GuideOct 12, 202614 min read
How to Evaluate an AI Visibility Platform: A Buyer’s Checklist
Every vendor deck lists the same engines and shows the same share-of-voice chart. Eight criteria separate them: four tell you which class of platform you are looking at, four tell you whether you can live with this particular vendor. Each comes with the question to ask on the demo call, what a good answer names, and a weight for your kind of team. The scorecard is a CSV you can take into the call.
- 50%of B2B buyers now start their journey in AI
- 89%of B2B buyers have adopted genAI
What an AI visibility platform is
An AI visibility platform runs the questions your buyers ask through AI answer engines such as ChatGPT, Claude, Gemini and Perplexity, records whether your brand is named, cited or recommended in each answer, and reports which competitors and sources appear instead. Some platforms stop at that report. Others diagnose the cause of each gap, build the fix and deliver it to your CMS as a draft for approval. BrandViz.AI calls the second kind a diagnose-and-ship platform, and the checklist below is built to tell the two apart on a demo call.
Eight criteria sort any vendor. The first four (engine coverage, query volume, diagnosis, fixes) tell you which class of platform you are looking at. The last four (CMS and Git delivery, approval, brand voice learning, pricing) tell you whether you can live with this one. Score each from 0 to 2, weight it for your team, and the number you walk out with is more useful than the deck.
Related: Tracking vs execution: the three classes of platform · What is GEO?
How to score a vendor on the call: 0, 1 or 2 per criterion, weighted for your team
You have three vendors shortlisted and the first demo is on Thursday. All three decks list the same engines, all three open on a share-of-voice chart, and two of them say they "fix" things. Decks name features, so the decks will not separate them; the call will, provided you ask questions whose good answers have to name a thing: an artefact, a platform, a person. That is what each criterion below is built to draw out.
There are three scores.
A 0 means the vendor cannot answer, or the answer is a feature name with nothing behind it.
A 1 means the vendor answers, and your team still does the work the answer describes.
A 2 means the answer names the artefact, the platform it lands in, or the person who approves it, and you could verify that on a trial.
Multiply each score by the weight for your kind of team (the buyer-type section suggests weights for an enterprise marketing lead, a lean B2B SaaS team, an agency and a govtech vendor), and the maximum is twice the sum of your weights. The scorecard below does the arithmetic.
Download the scorecard (CSV, opens in Sheets or Excel)1. Engine coverage: which engines, sampled from which market, how often
Engine coverage on an AI visibility platform means the set of answer engines it queries on your behalf, the market each query is run from, and how often each engine is re-run. Four engines sampled weekly from your buyer's country tell you more than eight engines listed on a pricing page and sampled once.
The engines disagree with each other more than most buyers expect. BrandViz.AI's own report for the week of 8 October 2026 ran the same question set through all four, 599 responses in total, and the share of answers naming the brand came back like this.
| Engine | Answers naming the brand |
|---|---|
| Claude | 12% |
| Perplexity | 9.3% |
| ChatGPT | 4% |
| Gemini | 2.7% |
A platform that blends those into one score would have reported about 7% and hidden the fact that one engine was naming the brand more than four times as often as another. Ask to see results per engine on the demo, with the answer text behind each number.
Market matters for the same reason. An answer generated for a buyer in the United States will list different vendors from one generated for a buyer in Australia, so ask whether a query runs as if from your market or whether global results are filtered by country after the fact. The first is coverage and the second is a label.
Which engines you need depends on where your buyers actually ask, and the honest answer differs by category. If Google's AI answers matter in yours, ask for them by name rather than assuming "multi-engine" includes them. BrandViz.AI tracks ChatGPT, Claude, Gemini and Perplexity, and would rather tell you that than list a longer set it samples thinly.
2. Query volume and markets: find out what the limit actually counts
A query limit on an AI visibility platform can count prompts, prompt-and-engine pairs, or individual runs, and for the same number on a pricing page the three differ by a factor of 4 to 20 in real coverage. Ask which unit applies, how many markets it covers, and how often the set is refreshed.
| What "50 queries" can mean | Real coverage | Where you see it |
|---|---|---|
| 50 prompts, every engine | 50 buyer questions, each run on every engine the platform covers, for one market. | BrandViz.AI Starter: 50 questions across ChatGPT, Claude, Gemini and Perplexity, refreshed monthly. |
| 50 prompt-engine pairs | 12 or 13 questions across four engines. A quarter of the coverage for the same headline number. | Common on dashboard-only tools priced per "tracked prompt". |
| 50 runs a month | A dozen questions, once each, on four engines, with nothing left over for a rescan. | Usually a usage cap on a free or entry tier. |
The set itself should be your buyers' questions, grouped by stage, rather than keywords with a question mark added. Rushana's line on calls is that the questions people are asking are the new keywords, and the practical test is whether you can write the prompts yourself and tag them by stage, or whether the platform hands you a generic set for your category.
Markets multiply the count. BrandViz.AI's Growth plan runs 150 questions across up to 3 markets, refreshed weekly; Starter runs 50 in one market, refreshed monthly. Whatever the numbers, the set needs to stay fixed between runs, because on BrandViz.AI customer data the engines typically take 4 to 12 weeks to reflect a shipped fix, and a platform that rotates prompts every month makes that movement unreadable.
3. Diagnosis: a cause you can act on, with the answer and its sources attached
An AI visibility platform diagnoses a gap when it names the cause behind each missing answer and shows the evidence: the full model response, the sources it drew on, and which of named, cited or recommended you failed on. A score with no answer text behind it can tell you that you are missing and nothing about why.
The named / cited / recommended split does real work. In the same report, BrandViz.AI's citation rate (10.2%) ran above its mention rate (7%), which means the engines were reading brandviz.ai as a source and then not saying the name. That is an attribution gap with a specific fix (organisation schema, consistent naming, entity resolution), and a platform reporting one blended number would never surface it. The causes we keep seeing fall into five classes, and a diagnosis should land each gap in one of them.
Missing content
No page on your site answers the question, so the engines answer from someone else's. The fix is a page, and a diagnosis should say which question it answers.
Technical
The page exists and the crawler cannot read it: blocked in robots.txt, rendered client-side with an empty HTML shell, or missing the schema that lets a model attach facts to your organisation.
Stale or wrong facts
The engines describe a product you retired or quote a price you never charged. The evidence is the answer text, and the fix is a page that states the current fact plainly enough to be quoted.
Cited but not named
Your site is in the sources and your brand is absent from the prose. An entity problem, visible only when citation and mention are reported separately.
Off-site
Review sites, directories and other people's pages carry the competitor and not you. The third of the three layers, and the one nobody can ship for you.
4. Fixes: the artefact that comes out, and the honest list of what cannot be shipped
A platform produces fixes when its output for a gap is a finished artefact: a JSON-LD block for a specific page, a rewritten metadata set, a full page in your voice. Recommendations, briefs and "optimisation suggestions" leave the work with your team, which puts the vendor in the dashboard-only class whatever the deck says.
The cost of the difference is time you can look up. SEOFOMO's State of Technical SEO survey (282 respondents, late 2024) found the most common time to get a technical recommendation implemented was 4 to 6 weeks when a developer was available, and the top reasons for delay were a lack of development resources and low prioritisation against other work. For a marketing team without a developer, a recommendation is a ticket that may never be raised, which is the whole argument in tracking vs execution.
Two follow-up questions sort the vendors who say they build. Ask what counts as one item, because a schema block on one page, an entity patch and a 1,500-word page are three very different units of work and some vendors count them the same. Then ask how many items a month, with that definition. BrandViz.AI ships around 8 a month on Starter and around 30 on Growth, and will tell you which kind each one was.
5. CMS and Git delivery: where the fix lands, and in what state
CMS integration on an AI visibility platform means the fix arrives inside the system your team already publishes from, in that system's own unpublished state: a draft post, an unpublished entry, a pull request. A PDF, a shared document or a "copy this snippet" panel is still a report, and the distance from it to live is a ticket in someone else's queue.
| Platform | The state a shipped fix should arrive in |
|---|---|
| WordPress | Draft post or page in wp-admin. |
| Webflow | Draft CMS item and staged SEO fields, unpublished until you publish the site. |
| HubSpot | Draft blog post or site page in the HubSpot editor. |
| Contentful | Unpublished entry in your chosen content type. |
| Git (GitHub) | A pull request on a brandviz-agent/ branch. |
The states BrandViz.AI delivers to. Ask any vendor for the equivalent row for your platform.
Teams that manage the site as a codebase should ask for a pull request rather than a CMS connector, because a PR goes through the review your engineers already run and touches only the files it names. Ask what the branch is called, whether the platform ever commits to main, and whether a schema fix arrives as a change to the page component or as a separate file your build will pick up. A vendor who cannot answer the third question has not shipped to a repository before.
6. Human-in-the-loop approval: who presses publish, and whether the vendor can
Human-in-the-loop approval means nothing the platform builds goes live until a named person on your team publishes or merges it, in your own CMS or repository, and the vendor has no path to publish around that step. Ask the question in that exact form, because "you can review everything" is also true of a tool that publishes first and lets you roll back.
"Who on our team presses publish, and is there any way for you to publish without them?"
The second half of the criterion is what happens when the draft is wrong. Ask where you request a revision, whether revisions cost anything, and whether the correction is remembered, which is criterion 7 arriving early. If the review happens in a second interface the vendor built, your editors learn a new tool to approve work that will be published from the old one, and the approvals pile up in the tool nobody opens. The better answer is that review happens in the CMS, where the draft already is.
For an enterprise or a government vendor, the location of approval is also the audit trail. If the change was made in your CMS by your editor, your CMS history already says who changed what and when, and nobody has to reconstruct it from a vendor's logs.
7. Brand voice learning: the second edit tells you more than the demo
Brand voice learning means the platform carries your corrections forward, so a positioning angle you fixed in one draft is not back in the next one. The test is simple: if you correct the same problem twice, the platform is not learning, and every draft will cost you the editing time it was meant to save.
Ask what, specifically, is stored from an edit. A good answer names categories: banned words, positioning angles, product names, pricing facts, the spelling convention. Then ask whether that store is per brand, which matters the moment you have a second product line or a second client, and whether you can watch it work on a trial by editing one draft and reading the next.
Edit one draft on the trial: change a product name, strike a word you never use, correct a price.
Send it back with no explanation beyond the edit itself.
Read the next draft for those three things. Anything that came back is your answer.
Terminology deserves its own question. When one page calls the product "AI visibility monitoring", another "brand tracking software" and a third "LLM brand monitoring", the engines synthesise a low-confidence picture of what you sell, and a platform generating pages at any volume will either enforce one name or multiply the problem. Ask whether it keeps a canonical term list and where it lives.
BrandViz.AI's Engine learns from the feedback on each item you approve or send back and carries it into the next batch, which is the behaviour the trial test above is checking for. Any vendor can say it; the second draft is where you find out.
8. Pricing transparency: the public price, and the price the engines think you charge
Pricing transparency on an AI visibility platform has two halves: the plan prices and limits sit on a public page, and the AI engines, when asked, repeat them correctly. Buyers check the first half and skip the second, and the second is the quickest test of whether a vendor can do for you what it is selling.
The test takes 2 minutes. Ask ChatGPT what the vendor costs, open the vendor's pricing page, and compare. If the engine is wrong about the vendor's own price, you have a live example of the problem you are buying a fix for, and the useful question on the call is how they would fix it for you.
On the first half, the questions are ordinary. Ask for the price per tier, what each limit counts (criterion 2 again), whether revisions and rescans are included, and what happens to the price when you add a market or a brand. A vendor who will only quote after a discovery call is not necessarily hiding anything, and is also not scoring a 2.
Weighting the checklist for your team: enterprise, lean SaaS, agency, govtech
The eight criteria are the same for everyone and the weights are not. A lean team scores execution and price highest because nobody else will do the work; an enterprise scores coverage and approval highest because the work is governed. The weights below are defaults, and the one line under each is the answer that should end the call.
Enterprise head of marketing
Weight engine coverage, query volume and markets, and approval at 3. You need the same buyer questions run per engine and per market, because a regional team will ask why the brand shows up in one country and not another, and a diagnosis that names a cause you can hand to the right team. Ask for the security questionnaire in the first week rather than the last, and accept an honest "not yet" on a certification over an "enterprise-grade" with nothing behind it. Multi-brand and custom security requirements sit on BrandViz.AI's Tailored plan rather than the self-serve tiers.
Walk away if: the vendor can publish to your site without a person on your team pressing the button.
Lean B2B SaaS team
Weight fixes, CMS delivery, brand voice learning and pricing at 3. You are the marketing team, or most of it, and engineering time arrives when a release slips rather than when a schema ticket is raised, so the only output that helps is one that lands in your CMS ready to read. Count items per month with the vendor's own definition, and run the pricing test from criterion 8 before the call so you are not the one being surprised. Coverage can sit at 2: four engines in one market is usually the right size for a team of one or two, and BrandViz.AI's Starter plan is built to that shape.
Walk away if: the artefact for a gap is a brief, because you already have a backlog.
Agency
Weight query volume and markets, fixes, brand voice learning and pricing at 3. The question that separates agency-ready platforms is whether each client is a separate workspace with its own query set, its own market and its own voice store, because a voice profile shared across clients will leak one client's positioning into another's drafts. Then ask how the query limit scales when you add a client, whether the report is something a client can read without you translating it, and what the per-client price is once you are past three. BrandViz.AI runs multi-brand agency setups on the Tailored plan, and the agencies page covers what that includes.
Walk away if: brand voice is a single global setting.
Govtech vendor
Weight approval, CMS delivery and diagnosis at 3. Your constraint is change control, and the only delivery model that stays inside it is one where the draft lands in the unpublished state your CMS already has and your existing approval step is the publish step. Diagnosis matters more here than elsewhere because a stale or wrong fact about a public-sector product is a procurement problem, and you need the evidence (the answer text, the source) to take to whoever owns the page. On certifications, ask for the list and accept the honest answer: BrandViz.AI has not completed IRAP, ISO 27001 or SOC 2 at this time, and a hard requirement on any of those is a Tailored plan conversation. The govtech page has the longer version.
Walk away if: the vendor describes an approval workflow that lives in their tool rather than yours.
The scorecard: eight rows, a weight column per buyer type, and a download
Score each row 0, 1 or 2 per vendor, multiply by the weight for your column, and total it. The maximum is twice the sum of your weights (36 for an enterprise or govtech column, 36 for a lean team, 40 for an agency). The CSV has the same rows with three vendor columns and a notes column, and nothing in it is specific to BrandViz.AI.
| # | Criterion | Ask on the call | A 2 names | Ent | Lean | Agy | Gov |
|---|---|---|---|---|---|---|---|
| 1 | Engine coverage | Which engines, sampled from which market, how often? | Each engine by name, the market setting, and the run cadence per engine. | 3 | 2 | 2 | 2 |
| 2 | Query volume and markets | What does the query limit count, and across how many markets? | The unit (prompts, prompt-engine pairs or runs), the market count and the refresh cadence. | 3 | 1 | 3 | 1 |
| 3 | Diagnoses causes | For a question we are missing from, what does it say the cause is? | A cause class (missing page, blocked crawler, stale fact, cited-not-named, off-site) with the answer text and sources attached. | 2 | 2 | 2 | 3 |
| 4 | Produces fixes | What artefact comes out for that gap? | A JSON-LD block, a metadata set or a full page draft. Not a brief, not a recommendation. | 2 | 3 | 3 | 2 |
| 5 | CMS and Git delivery | Where does the fix land, and in what state? | Your CMS by name and its unpublished state, or a pull request on a named branch. | 2 | 3 | 2 | 3 |
| 6 | Human-in-the-loop approval | Who presses publish, and can the vendor publish without us? | Your team, inside your CMS or repository. The vendor has no publish path. | 3 | 1 | 2 | 3 |
| 7 | Brand voice learning | What carries forward from an edit we make? | Named things: banned words, positioning angles, product names, pricing facts. Stored per brand. | 2 | 3 | 3 | 2 |
| 8 | Pricing transparency | What does it cost on a public page, and what does ChatGPT say it costs? | Tiers and limits on a page, and the model repeats them correctly. | 1 | 3 | 3 | 2 |
Weights: 1 = nice to have, 2 = matters, 3 = decides it. Change them; the defaults are a starting point.
Download the scorecard (CSV)Take it into Thursday's call. The vendor who scores highest will not be the one with the longest engine list, and in our experience it is usually the one whose answers kept naming your CMS.
Frequently asked questions
What should I look for when evaluating an AI visibility platform's ability to track prompt coverage across multiple AI engines?
Three things: what the query limit counts (prompts, prompt-engine pairs or runs), whether each prompt is run as if from your market, and whether the same set is re-run on a fixed cadence so movement is readable. Then check that results are reported per engine and split into named, cited and recommended. BrandViz.AI tracks ChatGPT, Claude, Gemini and Perplexity: 50 questions refreshed monthly for one market on Starter, 150 refreshed weekly across up to 3 markets on Growth.
What should an enterprise head of marketing look for when evaluating an AI visibility and content automation platform?
Weight engine coverage, market count and approval controls highest. An enterprise needs the same buyer questions run per engine and per market, a diagnosis that names the cause of each gap, and a delivery model where every change lands as a draft in the CMS the team already publishes from, so sign-off and audit history stay where they are. Ask for the security questionnaire early and accept an honest "not yet" over "enterprise-grade". Multi-brand and custom security requirements sit on BrandViz.AI’s Tailored plan.
What is the difference between an AI visibility platform that monitors and one that fixes?
A monitoring platform runs your buyer questions through the engines and reports where you are missing, who appears instead and which sources were cited. A diagnose-and-ship platform takes that same diagnosis, writes the schema, entity fixes and pages, and delivers them to your CMS as drafts for your team to approve. On the call, the difference shows up as a noun: a monitoring vendor names a feature, a diagnose-and-ship vendor names an artefact and the CMS it lands in.
Can an AI visibility platform ship content to a govtech company's CMS without bypassing internal IT approval?
Yes, if the fix arrives in the unpublished state your CMS already has and only your team can publish it. BrandViz.AI ships each item as a WordPress draft, an unpublished Contentful entry, a draft Webflow CMS item, a draft HubSpot page or a pull request in Git, and has no path to publish on your behalf. Your existing change-control step is the publish step. BrandViz.AI has not completed IRAP, ISO 27001 or SOC 2 at this time; a hard requirement on any of those is a Tailored plan conversation.
What should a marketing agency look for in an AI visibility platform for multiple clients?
Ask whether each client is a separate workspace with its own query set, its own market and its own brand voice, because a voice profile shared across clients will leak one client’s positioning into another’s drafts. Then check how the query limit scales with clients and markets, and whether pricing is predictable per client. BrandViz.AI runs single brands on Starter and Growth and multi-brand agency setups on the Tailored plan.
How much does an AI visibility platform cost?
Dashboard-only monitoring tools are the cheapest class, with entry plans under A$100 a month for a small prompt set on one or two engines. Diagnose-and-ship platforms cost more because the work is included. BrandViz.AI’s Starter plan is $200 AUD a month: diagnostics across ChatGPT, Claude, Gemini and Perplexity, 50 tracked queries refreshed monthly for one market, and around 8 Engine items shipped to your site each month. Growth is $1,000 AUD a month: 150 tracked queries refreshed weekly across up to 3 markets and around 30 items a month, landing pages included. Both self-serve plans start with 7 days free.
No credit card