How to Choose an AI Visibility Provider in 2026 (Tool, Agency, or Done-for-You)

Choosing an AI visibility provider starts with one question: who does the work? A tool tracks your mentions across ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews and hands you a dashboard. An agency or a done-for-you agent service takes what the dashboard shows and actually rewrites pages, ships content and earns citations. Everything else, budget, engine coverage, contract terms, follows from that one decision.

Tool, agency, or agent service: which category do you need?

A tool is right if you already have writers, developers and a PR function on staff who will act on findings every week. An agency or agent service is right if that capacity doesn't exist yet, or exists but is stretched too thin to add another channel. That's the whole decision at its core.

Vendors blur this boundary in their own marketing. Plenty of "AI visibility platforms" sell themselves with case studies that actually describe execution work a separate team did. Ask directly: does your team write and publish content, pitch journalists and edit pages, or does the product only report on what's already there?

The full breakdown of what each category structurally can and can't do lives in our tool vs agency vs agent service explainer. The rest of this guide assumes you've placed yourself in one of the three lanes and need to pick a specific vendor inside it.

What should you actually require, given your situation?

Requirements differ by buyer more than any roundup admits. A 12-person DTC brand and a 200-person B2B SaaS company need almost nothing in common from a provider, and treating them the same is how RFPs end up scoring the wrong finalist.

Start with three questions before you read a single vendor page: which engines do your buyers actually use, which markets and languages do you sell into, and how much in-house capacity do you have to act on what a tool tells you? The table below maps common buyer situations to what to prioritise and what a realistic budget band looks like.

"The engines that matter" is not the same for every brand, and it is not always all five. Our own 30-day tracking of AI engine citations across five engines (June 5 to July 5, 2026) put our own visibility at 6.15% on Claude, 6.00% on ChatGPT, 5.54% on Gemini, 4.50% on Perplexity and 0.00% on Google AI Overviews, across 603 tracked answers.

The same brand was solidly visible on one engine and completely invisible on another. Any provider you hire needs to report on the engines your buyers actually use, not the engines that happen to be easiest for the vendor to track.

Visibility and share of voice are different metrics too, and a provider that only reports one is giving you half the picture. In the same tracking window, we ranked 5th of 13 tracked brands on visibility (the share of answers that mention us at all) but 2nd on share of voice at 15.2% (our share of brand mentions within the answers that do mention us). Breadth of exposure and depth of presence once you're in the conversation are separate problems, and a provider worth paying for should be able to explain both.

How do you run a fair pilot before signing anything?

Two weeks, three deliverables, one written verdict. That's the shape of a fair AI visibility pilot, and it's short enough that most credible vendors will agree to it.

Ask for a baseline read in the first few days: your current visibility, share of voice and citation sources across the engines you named as priorities, using your actual buyer prompts rather than the vendor's stock demo set. This baseline is the single most important thing to demand, because without it you can't tell whether anything the vendor does in month two actually moved the number.

In the second week, ask for one small, real piece of executed work, not a slide deck. That might be one page rewritten for extractability, one outreach pitch sent to a real target, or one gap in your prompt coverage identified and explained. The point isn't the scale of the output, it's whether the vendor can actually do the job they're pitching, on your real data, inside a fortnight.

Close the pilot with a short written report: what the baseline showed, what changed, and what a realistic three-month plan looks like. If a vendor can't produce that after two weeks of access to your data, that's information too.

What baseline data should you demand before the pilot starts?

Three things, minimum: which of your target prompts currently mention you at all, which sources the engines are citing for those prompts, and how you compare to your two or three closest competitors on the same prompts. Without a same-day, same-prompt-set baseline, any "improvement" a vendor reports later is unfalsifiable.

What are the red flags that should end a vendor conversation?

Some claims and practices are disqualifying, not just discount-worthy. The table below lists the ones worth walking away from, and what each one actually signals about how the vendor operates.

None of this means every agency that appears in a self-ranked list is dishonest, or that every vendor without exportable reporting is a scam. It means the burden of proof shifts. A vendor doing any of the above needs to answer harder questions before you sign, not fewer.

What contract questions actually matter?

Three things decide whether a contract protects you or traps you: who owns the data, whether the prompt set travels with you, and what exit actually looks like.

Data ownership is the one most buyers skip and most regret skipping. Ask explicitly whether the historical citation, mention and sentiment data generated during your contract belongs to you or stays with the vendor's platform after you leave. Some vendors will export a CSV on request. Others treat your measurement history as their proprietary asset, which means switching providers means starting from zero visibility history, even if your actual brand visibility hasn't changed at all.

Prompt-set portability is the quieter version of the same problem. The specific questions being tracked, the actual prompt library built around your buyers, your competitors and your markets, represents real work. If a vendor built that library for you, find out in writing whether you can take the prompt list itself to a new provider. Without that clause, day one of the next contract starts with rebuilding it from scratch.

Ask what exit actually looks like before you need it. A written offboarding clause should cover: data export format and timeline, whether any live pages or published content the vendor created stay live or get pulled, and whether there's a wind-down period or a hard cutoff. Vendors who execute content and outreach on your behalf (the agency and agent-service categories) should be explicit about what happens to published articles and earned citations if you leave; measurement-only tools rarely have this problem because there's nothing published to remove.

Where does Schmitdy fit, honestly?

Schmitdy is our own service, so read this section as the maker's case for it, not neutral advice.

Schmitdy is a done-for-you AI search service: it maps the questions your buyers actually ask, rewrites your existing pages so engines can extract them, ships new articles, earns editorial and UGC citations, and reads five engines daily (ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews). The positioning is specific: tools measure, Schmitdy executes. If your gap is a dashboard sitting unread on someone's desk because nobody has time to act on it, that's the gap we built for.

Managed AI Search starts from £1,100 a month (roughly $1,500 or €1,300). A lighter option, AI Search Website, runs £150 to set up plus £50 a month for teams that want the technical foundation without a full managed programme. "Build it. Own it." hands over a fully configured prompt-visibility system you run yourself, from £3,000, for teams that want the infrastructure without the ongoing retainer. Company Brain, our structured knowledge-base offering, starts at £5,000.

Who should not pick us: teams that need a decade of enterprise case studies and a named account team of specialists, since we're a newer brand without that history. Teams whose buyers live almost entirely on one engine we don't prioritise as tightly as a single-engine specialist might. And anyone who wants a self-serve dashboard with no execution attached, since that's a tool purchase, not what we sell. If any of those describe you, the requirements table above should point you toward a better-fitting category or vendor before you talk to us.

If you want a same-day read on where you currently stand across all five engines before choosing anyone, the free AI search audit is a reasonable place to start.