Choosing an AI visibility tool in 2026: eight questions to ask, and how Tarnua Visibility answers them

Some buyers now ask ChatGPT, Gemini, Claude or Perplexity before they open Google. If those answers don’t name you, you’re invisible at the exact moment someone is ready to choose. A new category of tools, AI visibility (or GEO) platforms, measures whether that’s happening to your brand.

We build one of them, so we won’t pretend to be a neutral reviewer. What we can do is list the questions we’d ask any tool in this category before paying for it, and answer each one for Tarnua Visibility with something you can check. We don’t describe or rate other vendors here: their features and prices are theirs to change, and we only publish claims we can prove.

1. Which AI surfaces does it measure, and which of them search the web?

Two different things hide behind “AI visibility”. An engine that answers from what its model already knows can only mention you. An engine that searches the web as it answers can cite you: a real link to your domain inside the answer. A tool should tell you which is which, per engine, on every run.

Tarnua Visibility: four chat engines (ChatGPT, Gemini, Claude and Perplexity) plus Google AI Overview as a sampled fifth surface. Every report labels each engine “live” or “model” for that run, and only a live engine can earn a citation. We switch on an engine’s web-search mode only where its provider’s terms allow the analysis we run on the results. Where they don’t, that engine answers from model knowledge, and the report says so rather than quietly using it.

2. Can you read the actual answer behind every number?

A score you can’t trace is an opinion. Ask whether every data point opens the real AI response it came from, with the date and the exact model that produced it.

Tarnua Visibility: every cell in a report run today opens the full answer: dated, quoted, with the model named and sensitive claims flagged. AI statements are presented as AI statements, never as facts about your brand, and never as facts about your competitors either. A report will tell you that an engine named a rival, and nothing else about them: not their size, ownership, clients or press. We don’t publish a verdict on a company we didn’t audit.

3. Who writes the prompts, and how many are branded?

A prompt that contains your brand name measures whether AI knows your name. A prompt a buyer would actually type measures whether AI recommends you. Ask what share of the prompts are non-branded, and whether you control them.

Tarnua Visibility: we read your page and generate the buyer prompts, about 95% non-branded unless you say otherwise, seeded from real search-demand signals. You steer them in plain English (“no branded prompts”, “only worldwide”, “consider 50% pricing topics”), and those instructions are re-checked in code against every generated prompt, with the failures sent back for rewrite. Or paste your own.

4. Are the answers asked from your buyer’s market?

The answer given to someone in Dublin isn’t necessarily the answer given to someone in Berlin. If a tool shows one answer under three flags, it has measured one market.

Tarnua Visibility: up to five target markets per report, with per-market prompts asked from each market, so a per-country column is a separate measurement.

5. Is run two comparable to run one?

Models change, providers change, and a tool’s own settings change. If the score moves, you need to know whether your visibility moved or the instrument did.

Tarnua Visibility: every prompt is a fresh API call with no chat history or personalisation, scored by the same published rubric. Each run records the models it used, whether each engine searched the web, and the answer settings it ran under. The report shows a banner when any of those differ from the run you’re comparing against. An unavailable engine is excluded, never scored as a zero.

6. Where does your data go?

Your brand name, your pages and your prompt set leave the tool and reach AI providers. Ask for the list, the regions, and whether your inputs can be used to train models.

Tarnua Visibility: SYNECT LIMITED, trading as Tarnua, is a Dublin company. Our privacy policy names every sub-processor and region in plain English, says which engines send a search query to the open web, and names the two providers that receive your inputs on every report whichever engines you enable. Every audit has a hard spend cap, and you can export or delete everything from Settings.

7. What does it claim about your competitors, and can it prove it?

Be wary of any tool that tells you a rival’s strategy, budget or customers from an AI answer. The answer is a model’s statement, not a fact, and repeating it as a fact is how a report becomes a liability.

Tarnua Visibility: we report which brands the engines named and cited ahead of you, from the answers themselves, and never add a verdict on the rival’s business. The same rule applies to this page.

8. What does it cost, and what do you get for it?

Look for a public price, whether it excludes VAT, how usage is metered, and what happens when you cancel.

Tarnua Visibility: plans from €49/month excluding VAT, usage metered in credits (one credit per prompt per engine) and the full list on the pricing section. Early access: workspaces are approved by hand before anything runs.

What Tarnua Visibility doesn’t do yet

• No prompt-volume demand data. We use demand signals to shape the prompts; we don’t show you search volumes.
• No API or MCP access.
• No Copilot, Grok, DeepSeek or Google AI Mode surfaces.
• No automated “agents” that change your site. The action list is written for a person to act on.

If you need one of those today, another tool may fit better. We’d rather you pick the right one than pick us for the wrong job.

The short version

See exactly what a report looks like (no signup) at the sample report, or read the full methodology.

Rewritten on 6 September 2026. The original version (31 August 2026) compared five other vendors’ features and prices. That comparison has been removed: we publish only claims about our own product that you can check. Spotted something wrong? Email hello@synectgroup.com and we’ll correct it.