AEO vendors have learned to make monitoring sound like execution. If an agency cannot tell a prompt tracker from a publishing workflow, it will resell a dashboard and call it a retainer.

I know the trick from Google Ads. A feature flags a problem, a sales page describes the outcome, and the operator is left to do the work between them. The terms below have real meanings. They also have stretched meanings that make for a much better demo. I use the same test for each: what happens after the tool notices something?

Tracked prompt

A tracked prompt is a buyer question a vendor runs on a schedule across AI search surfaces, recording whether and how your brand appears.

The prompt cap is the throttle on the product. Consider Otterly Lite at $29 for 15 prompts, $189 for 100, and $489 for 400. Three clients with 40 questions each push you past the 100-prompt tier before anyone has changed a page.

A home-services client can generate 60 high-intent questions without trying: cost, near me, best, versus, reviews, emergency and warranty, multiplied by location. Track only 50 and you are choosing which blind spots to keep. What the deck calls scale, I call a meter running.

Ask on the demo: What happens on prompt 51, and what does 500 prompts across 20 domains cost for a year? If the math kills your margin as you add clients, you are buying a sampling habit, not execution.

Share of voice

Share of voice is your share of mentions or citations within the set of AI answers a tool tracks.

The calculation looks clean. The denominator is where things get slippery. One vendor counts the five competitors you entered; another counts every brand named in the answers. Those scores can tell different stories about the same week, even when both are presented as your share of tracked mentions or citations.

I used to tell clients impression share meant winning. I was wrong in the same way: a high share of a tiny, friendly set can still miss the buyer. Ask whether the score measures mentions or citations, and who sits in the denominator. If the vendor cannot show those parts separately, the score is a mood ring.

Citation vs mention vs recommendation

A citation links your URL as evidence, a mention names your brand, and a recommendation tells the buyer to choose you.

Vendors fold all three into visibility because the combined number looks better. But brands are three times more likely to be cited alone than cited and mentioned together. You can supply the footnote for an answer that recommends your competitor.

I have seen the PPC version for years: an assisting click gets treated like the closing click. Helping the answer is not the same as winning the buyer. Ask for a prompt where the client is cited but a competitor is recommended. Then ask what the vendor would change. A tool that only counts appearances cannot answer the second question.

Citation building

Citation building is the work of earning independent pages that support a claim an AI answer might use.

That could mean a review, a listicle, a data roundup or a relevant forum discussion. The output is a live page, not a list of places where the client should appear. I learned the distinction doing link building for ecommerce: a report saying you need 40 citations is not 40 citations.

When I assess the work, I look for three things:

Ask for three live citations earned last month for a client in your niche. If the vendor shows a prospect list or an export of existing mentions, it is selling research as results.

Content execution vs content briefs

A content brief tells someone what to make; content execution gets the page written, prepared and published.

A brief can identify the buyer question, suggest the structure and recommend schema. Someone still has to write the copy, add the links, check how the page renders and hit publish. That is why I draw a hard line between a useful instruction and completed work.

The distinction appears in Rankability’s separation of monitoring tools from a fuller workflow: it notes that Profound stops at insight without a content-creation layer and Peec has no content optimization. Early PPC scripts taught me the same lesson. A flag for wasted spend felt productive until I checked whether anyone had changed the account.

Ask who writes the page, adds the schema, publishes it and shows you the URL. If the answer is your copywriter, your developer and your spare afternoon, file it under briefs and price it that way.

Auto-publish to CMS

Auto-publish to CMS means a completed page goes live on the client’s site without someone manually moving it there.

A WordPress plugin that creates a draft, a Zapier step that sends HTML to staging and an export you paste yourself are different workflows. All may be useful. None is the same as a live, checked page. For an agency running 20 retainers, the time between drafted and published is where margin goes to die.

My test is boring on purpose. Ask the vendor to publish a test page to your holding site, with a title, body copy, internal links and schema, then show the live URL. Check what the client must approve and who verifies the result. Publishing does not itself guarantee indexing. If the process ends in an IT ticket, call it CMS-assisted drafting and staff it that way.

Knowledge graph management

Knowledge graph management is the ongoing work of making a brand’s identity and relationships consistent across its site and relevant third-party records.

It sounds like a dial you turn. In practice, the work may include appropriate entity listings, JSON-LD markup, consistent descriptions and internal links that use the same names. This guide describes the work in terms of claiming, markup and internal linking, not one-click calibration.

The companion rule I use is simpler: keep one brand name and a consistent one-sentence description across the places you control. Ask for the diff: what listing, markup, link or description changed last month? A score is a thermometer. A corrected record is work.

Technical fixes

A technical fix changes the site so a bot can read content it previously could not.

Crawlable only means a bot could fetch a page. It does not prove the bot received useful content. A page may look fine to a person yet return near-blank HTML to an AI crawler because of client-side rendering, blocked scripts or CDN rules. A dashboard can flag that problem without fixing any of it.

I treat a list of 140 issues with zero changes shipped like a PPC audit that finds wasted spend and changes no bids. The fix needs a before-and-after fetch, not a greener score. Ask for one URL bots could not read, the markup or server-rule change, and the fresh fetch that verifies it. If all the evidence lives inside the vendor dashboard, budget for your own technical check.

Edge proxy

An edge proxy handles requests at the network edge and can serve AI crawlers a more readable version of a site’s pages.

The appeal is obvious when an existing site has many pages the crawlers struggle to read: clean HTML and schema without rebuilding the CMS. But edge setup can also mean a DNS change, a worker and a prayer that checkout keeps working. I like the mechanism when the guardrails are explicit: versioned rules, a rollback path, and exclusions for cart and logged-in pages.

Ask who owns the DNS record, what latency the setup adds, and how the bypass works. Then ask what a crawler and an ordinary customer receive if the proxy fails. If every domain needs a senior DevOps call, do not buy it as a no-developer installation.

No-developer installation

No-developer installation means an agency can connect a client site and start using the product without a code deployment or an IT ticket.

The stretched version means the vendor does not need a developer while you still paste scripts, approve Cloudflare access and debug a staging theme. I budget that gap in hours. A 20-minute connection across 30 clients is 10 hours; a two-hour DNS and QA loop across 30 is 60. That is not a small-print difference.

Time a fresh setup on the demo. Ask the vendor to onboard a subdomain while you watch, including permissions and checks. If step three is “call your developer,” price the tool plus the developer.

White-label

White-label means the client experiences the agency’s service without encountering the vendor’s brand in the workflow.

A logo on a PDF is not the whole workflow. Resellable fulfillment should keep the agency’s logo, domain and emails client-facing, rather than turning every report into a referral. The details are where the vendor name tends to leak: brand colors, a custom reporting domain, footer contact information and automated emails.

Ask for a live client report, a notification email and the portal URL exactly as your client would see them. If any carries the vendor’s name, you have co-branding you will have to explain. More important, ask what actually ships under your name. A branded dashboard is still a dashboard.

Flat-rate per domain

Flat-rate per domain means a set price for each client site, with defined work and usage included.

The phrase gets stretched when the headline price is flat but the meter moves into the footnotes: more prompts, more AI engines or more frequent refreshes all cost extra. A $99-per-month per-domain add-on such as Semrush’s may be easy to understand for one site; across a client roster, add-ons change the retainer math.

Say you are pitching a $1,500 AEO retainer alongside $20,000 a month in media spend. Do not let the AEO tool’s base price stand in for its total cost. Ask for 300 prompts and daily refreshes on one domain in writing, with every add-on named. Flat only counts for the workload you intend to sell.

Multi-domain pricing and seat traps

Multi-domain pricing is what the bill becomes as you add client sites; a seat trap is the extra cost or friction when more people need access.

This is where agency math breaks. A tool advertised at $99 Starter or $399 Growth with one workspace per account can look different once account managers, strategists and clients need separate access. New workspaces and logins cost time even before an enterprise quote arrives.

I price it as I used to price PPC seats: total annual cost for 25 domains, 500 prompts each, three seats and weekly reporting. Ask for that number on the call. If the vendor will quote only one domain and one seat, you know how much of your agency workflow the headline price covers. I keep this breakdown of flat-rate AEO pricing per domain open during demos.

Five questions I take to every demo

The individual tests matter, but these five force the conversation off the spec sheet:

  1. Show me last week’s action log for a real client: what shipped, why, and what changed in the tracked prompts afterward?
  2. What is the written annual price for 500 prompts across 20 domains, with seats included?
  3. Can you publish a test page to my holding site now and show me the live URL and schema?
  4. Can you forward a report, email and portal link exactly as my client would see them?
  5. Can you show a prompt where the client is cited but a competitor is recommended, then show what you changed in response?

I keep the AEO glossary beside the demo for the plain definitions. A vendor that dodges the action log is asking you to trust the label instead of the work.

AI visibility score

An AI visibility score is a summary number a vendor assigns to a brand’s appearances in a chosen set of AI answers.

If I could delete one phrase from AEO sales pages, it would be AI visibility score. It borrows the authority of a number to conceal choices about prompt samples, denominators and whether a citation counts like a recommendation. I watched impression share do a version of that job in PPC: a clean percentage could move when you changed the settings, not the market.

Retire the score. Keep the two counts underneath it: how many buyer prompts recommend you, and how many cite you as proof? Tracking earns its keep when it determines what ships next. Execution earns it when a page, listing or code change goes live under your agency’s name. Everything else is a dashboard you will have to explain to a client who pays for outcomes. Ask for the log first.

Frequently asked questions

How many tracked prompts do I actually need for an AEO monitoring tool?

Plan for each client's real question set, not a token sample. A single home-services client can generate 60 high-intent questions once you combine terms like cost, near me, best, versus, reviews, emergency and warranty across locations. Three clients with 40 questions each can push you past a 100-prompt tier before anyone has changed a page.

Why do different AEO tools give such different share of voice scores for the same brand?

Because the denominator differs. One vendor counts only the competitors you entered, while another counts every brand named in the tracked answers, so the same week can produce very different scores. Ask whether the score measures mentions or citations and who sits in the denominator; if a vendor cannot show those parts separately, the number is not reliable.

What is the difference between a citation, a mention and a recommendation in AI answers?

A citation links your URL as evidence, a mention names your brand, and a recommendation tells the buyer to choose you. Vendors often fold all three into a single visibility number, but a brand can be cited as the footnote in an answer that recommends a competitor. Ask to see a prompt where the client is cited but a competitor is recommended.

How do I know if a vendor is really doing citation building or just selling research?

Ask for three live citations earned last month for a client in your niche. Real citation building produces live independent pages, such as reviews, listicles, data roundups or forum discussions, with quotable passages and credible context. If the vendor shows a prospect list or an export of existing mentions, it is selling research as results.

Is a content brief the same as content execution in AEO software?

No. A brief identifies the buyer question, suggests structure and recommends schema, but someone still has to write the copy, add links, check rendering and hit publish. Ask who writes the page, adds the schema, publishes it and shows you the URL. If the work falls back on your own copywriter and developer, it is a brief and should be priced as one.

What does auto-publish to CMS actually mean for an AEO tool?

It means a completed page goes live on the client's site without someone manually moving it there. A WordPress draft, a Zapier step to staging, or an HTML export you paste yourself are different workflows and not the same as a live, checked page. Ask the vendor to publish a test page with title, body copy, internal links and schema, then show the live URL.

How can I verify an AEO vendor actually fixed a technical crawlability problem?

Ask for one URL bots could not read, the markup or server-rule change made, and a fresh fetch that verifies the bot now receives the content. Crawlable only means a bot could fetch the page, not that it got useful content, since client-side rendering, blocked scripts or CDN rules can return near-blank HTML. If all the evidence lives inside the vendor dashboard, budget for your own technical check.

Is flat-rate per domain AEO pricing really flat?

Only for the workload defined in writing. Headline prices can be flat while extra prompts, extra AI engines or more frequent refreshes cost extra, which changes the retainer math across a client roster. Ask for the workload you intend to sell, such as 300 prompts and daily refreshes on one domain, in writing with every add-on named.