You've decided your business needs to be visible when customers ask ChatGPT, Gemini or Google's AI who to use.
Now you're choosing a platform, and the product pages all say much the same thing.
Five criteria separate them in practice: which engines get sampled, how often and how reliably, whether the work ends at a report or at a published page, whose voice the content is in, and what the vendor will honestly promise.
Each section below stands on its own, so you can use it as a checklist against any provider, including us.
Which answer engines it samples
Different engines read different pages. Kevin Indig's "The Consensus Gap" (Growth Memo, 11 May 2026) found that only 2.37% of cited URLs appear across all three engines for the same prompt. A platform that watches one engine tells you about one engine, and your buyers aren't all asking the same assistant.
Good looks like a named list of engines, including Google's own AI surfaces as well as the standalone assistants, and a clear statement of which are included on which plan. Ask what happens when the engines disagree. A platform that averages them into one score has hidden the most useful information it collected.
We sample up to six: ChatGPT, Gemini, Perplexity, Google AI Overviews, Claude and Google AI Mode, with the number depending on your plan.
How often it checks, and how it asks
One snapshot isn't a measurement. A University of St. Gallen study (Schulte, Bleeker and Kaufmann, arXiv:2604.07585) found that only 34–42% of cited sources overlap between consecutive days. If the sources move that much, a single check tells you what happened on one afternoon.
Wording matters as much as timing. The same buyer question can be phrased several ways and return different sets of businesses. Good looks like repeated checks over time, more than one phrasing per question, and a rotation that covers your whole question set instead of the same few questions every time.
There's a trade-off here. More frequent checks give you a finer picture of movement, and they also show you how noisy the answers are. A platform that reports a rank to one decimal place is claiming a precision the engines don't have. Judge your position by the pattern across samples.
We ask every tracked question in three phrasings, rotated weekly, and we check the least-recently-sampled questions first so the whole set is covered across the month. On Basic and Growth we re-check core questions weekly.
Whether it ends at a report or at a published page
A dashboard that shows you're missing from an answer is useful for about a day. Then someone has to decide what to write, write it, and put it somewhere the engines can read. If that part is left to you, the platform has measured the problem and handed it back.
Good looks like a path from "you're missing from this question" to a draft, to a published page, with the publishing connected to where you already post. It also helps to know what you're writing toward.
A Princeton and IIT Delhi paper presented at KDD 2024 (Aggarwal et al., arXiv:2311.09735) tested content changes and found that adding statistics raised visibility by about 33%, adding quotations by about 41% and citing sources by about 28%, while keyword stuffing lowered it by 8%. Specific, sourced writing helped, and padding hurt.
The source mix matters too. A University of Toronto study of consumer-electronics queries (Chen, Wang, Chen and Koudas, arXiv:2601.16858) found that for "who should I use" style queries, 59–86% of cited sources were third-party. For transactional queries, 52–68% were brand-owned.
If you sell expertise, the questions that matter to you are mostly the first kind. The platform should help you earn mentions beyond your own site as well as publish on it.
We publish to LinkedIn, WordPress, Ghost and your own site blog, and we draft in eight language variants: UK English, US English, French, Spanish, German, Portuguese, Dutch and Italian.
Whose voice the content is in
AI-drafted content that sounds like everyone's content gives an engine no reason to pick you. Readers can tell too, and so can the people inside your firm who are asked to publish it. That second group decides whether a rollout lasts, because if the draft doesn't sound like them, they stop using it.
Good looks like drafts that start from each person's own voice, and a clear rule about their words. Check what the platform does with a claim attributed to a colleague. If it can put that claim into a draft without the person seeing it, you have a trust problem waiting to happen.
We draft in each team member's voice, and a colleague's claim can't be used in a draft until that colleague has approved the wording themselves.
How the question map grows
You'll never list every question your buyers ask. Platforms handle this in one of two ways: they track a fixed list, or they keep adding questions. Automatic additions can bury you in noise and make the numbers hard to compare month to month.
Good looks like a map that grows from your results and a person who approves each change. The questions adjacent to ones you already win are the likeliest next gains, because the engines have already shown they'll cite you on the topic.
When you get cited for a question, we propose a handful of adjacent questions to track next. Nothing is added until a person approves it, because the map grows as you win.
What the vendor will promise
No one controls what an AI recommends. Two people asking similar questions can get different answers from the same tool, and some engines search the web live while others lean on training data. A vendor who promises you'll be named is promising something outside their control.
Good looks like precise claims about what's measured and what's been changed, a plain distinction between being discoverable and being recommendable, and no guarantee of an outcome. Better information improves your chance of being considered, but it doesn't settle the result.
We measure where you appear, identify gaps and help you address them. We don't promise that an AI will recommend you.
Who this kind of platform is wrong for
An AI visibility platform fits a business whose buyers ask "who should I use?" and then pick a name they've come to trust. That means consultancies, agencies, advisers and specialist B2B firms.
It's wrong for a business that competes mainly on price or depends on high-street footfall. It's also wrong for a business selling commodity goods, where the buyer has already decided what to buy and only needs the cheapest listing, and for anyone who wants a guaranteed position, because no honest platform in this category can sell you one.
Who we're the right choice for
Ghostart is built for founder-led, expertise-led firms where the founder is the brand, and for marketing leads in small firms who need several people publishing in their own voices. If you want measurement, a plan, drafting and publishing in one system, with the engines re-checked so you can see what changed, that's what we do at ghostart.io.