What should an agency managing 10+ client brands look for in an AI visibility tool?

An agency tool has two jobs. It has to measure the right things, and it has to scale across a roster without breaking your workflow. Put these six requirements on your vendor scorecard:

• Brand capacity that fits your roster. Check the brand, keyword and audit caps on each plan, then multiply by your client count. A low brand cap means several logins, several invoices and scattered history.
• A separate prompt set per client. Each client has different buyers asking different questions. A shared prompt list blurs clients together.
• Per-engine results. ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews draw on different indexes and behave differently. Each one needs its own line in the report.
• Like-sized competitor benchmarks. A regional client compared with a household name produces a chart about brand size, not about your agency's work.
• Access checks per client domain. Robots.txt rules, firewall blocks and JavaScript-only content are the problems with the strongest evidence behind them, and the client can fix them.
• Accuracy findings. Wrong prices and outdated product names are what clients complain about, and the client can correct its own pages.

One thing no tool can give you is a client's position inside an AI answer. No engine publishes that data. The tools worth paying for say plainly which numbers are measurements and which are their own model. If you resell under your own brand, our guide to white-label AI SEO covers the extra checks.

A single cross-engine score hides the differences that matter most. In Otterly's 2026 experiment with two AI-generated test sites, Microsoft Copilot produced 2,718 of the 3,934 AI citations tracked for one site (69%). ChatGPT, Gemini, AI Overviews and AI Mode almost never cited it. Bing kept the site indexed after Google removed it, and Copilot grounds on Bing.

Blend those engines together and the client sees a middling number. The real story is strong on one engine and near zero on four. That story points to a fix: index status per engine. The blended number points to nothing.

Even within one company, surfaces differ. Google says AI Overviews and AI Mode "may use different models and techniques" and can return different links. Retrieval also fires per query at the model's discretion, so any single score averages answers given with and without live search.

Ask every vendor to show results engine by engine, with retrieval state disclosed where possible. Our page on whether one dashboard can track every AI engine covers what a combined view can and cannot do.

Prompt sets are samples, not the real input. OpenAI's help centre says ChatGPT search rewrites a user's question into one or more targeted queries before it searches. A tracker only sees the answers to the prompts it chose.

The gap can be large. In the same Otterly test, Microsoft's own AI Performance report credited one site with 42,817 citations, against 3,934 logged by Otterly's prompt-based tracker. The time windows may not match, so it isn't an exact ratio. It does show that an engine's own count sees more than any prompt sample. Bing Webmaster Tools is the only place that shows the grounding queries an AI actually ran, so set it up for each client using our Bing Webmaster Tools playbook.

Benchmarks need the same care. Citations concentrate on big names: a study of AI Search Arena data from March to May 2025 (arXiv 2507.05301) found the top 20 news sources accounted for 67.3% of citations for OpenAI models. Brand prominence is the largest uncontrolled factor in this field. Pick three to five competitors of similar size and market for each client.

Access is where the evidence is strongest, and it varies client by client. Check three layers on each domain.

Robots.txt, per bot

OpenAI's crawler page says each bot setting "is independent of the others". GPTBot governs training. OAI-SearchBot governs ChatGPT search. A client can block one and allow the other:

User-agent: GPTBot
Disallow: /

User-agent: OAI-SearchBot
Allow: /

The most common error is a blanket AI block that quietly removes ChatGPT search eligibility.

Firewall and CDN

OpenAI says eligibility needs OAI-SearchBot allowed and the host or CDN letting its published IP addresses through. A firewall can return 403 to a bot that robots.txt allows. Since 15 September 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers. A client can lose access without editing anything. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.

Rendering

Vercel's crawler study (December 2024) found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. Googlebot renders. Content that appears only after JavaScript runs may be invisible to most AI crawlers. See client-side vs server-side rendering for AI crawlers.

Split every client report into two parts.

What the client controls. Crawler access per bot, firewall results, pages whose key facts appear only after JavaScript, accidental noindex or nosnippet tags, index status in Google Search Console and Bing Webmaster Tools checked separately, and factual errors with the page that should state the correct fact. Each item has an owner and a fix.

What the client observes. Citations and mentions per engine, against like-sized competitors, labelled as a sample of chosen prompts. Present these as observations, never as targets you guarantee. Google's own AI report in Search Console shows impressions only, with no clicks and no position.

Drop anything the evidence contradicts: E-E-A-T scores, domain authority targets and AI rank tracking. Our agency client reporting playbook gives a template, and tracking tools vs fixing tools explains why the fix list matters more than the chart.

If you're weighing 99Visibility, note its largest plan, Pro at $149 a month, covers 5 brands, so a 10+ brand roster needs planning. We cover that fit in is 99Visibility a good fit for agencies.