The most complete AI visibility snapshot (we could find) is now free

From today, everyone can run a full Heralded Snapshot for free, with no account.

It is also the most complete AI visibility snapshot we could find. We checked nearly 50 one-off scans, free and paid, before putting that sentence on our website. Ours included, which we realise is exactly what every vendor says right before ominously winning its own comparison. So we published the whole thing, with dates, including the places where somebody else wins.

The short version. We compared seven things a snapshot can give you. The Snapshot does all seven. The next best free alternative does two.

If you lead marketing at a European B2B tech company, it takes 10 minutes and it answers a question your board has been asking for the last few months. When your buyers ask an AI engine what to buy, does it name you? And if it doesn’t, who does it name instead, and why?

Word of mouth is a chatbot now

Your buyers used to ask a colleague, read a few reviews and search Google. Now a lot of them ask ChatGPT first. G2 found that 51% of B2B software buyers start their research in an AI chatbot more often than in Google, up from 29% a year earlier. 6sense puts the share of B2B buyers who use an AI model somewhere in the buying process at 94%.

They use it to research and compare, which is exactly where a shortlist gets made. By the time a buyer reaches your website, the engine has already told them who the serious options are, what each one is good at and which one to try first.

The engines form those opinions from everything written about you. Your site, your competitors’ comparison pages, review platforms, listicles, forum threads from 2022. You don’t control what they say. You can change what they read, but only once you know what they are reading.

And most companies don’t. Your analytics sees the visit that comes afterwards. It never sees the answer that put three of your competitors on the list and left you off. Corporate Ink found that 72% of B2B tech marketers have seen AI describe their company wrong. Those are the ones who happened to look.

One answer is an anecdote

So people check by hand. They type “best tool for X” into ChatGPT, spot their name or don’t, and draw a conclusion.

The engines disagree with each other, and they disagree with themselves. When I ran the first version of this scan on Notion in June, ChatGPT called Notion the leading option on 3 of the 10 questions it asked. Gemini did it on none. Same questions, same week. Ask ChatGPT the same question twice and you can get two different lists, both delivered with complete confidence.

That is why a Snapshot exists. To know where you stand you need the right questions your buyers actually ask, put to every engine they use, more than once, with the sources behind every answer. Nobody does that by hand for long.

Most free checks don’t get there either. Some never ask an engine anything and grade your HTML instead. The ones that do ask usually run a preset list of generic questions, on a few engines, once.

What is a Snapshot

It checks what six AI engines say about your brand, reads the cited sources and your own site, and gives you a report on what to fix.

Why we’re confident to call it the most complete

The comparison lays it out row by row. Plenty of free tools do one or two of those rows well. None of them did all of them in one scan. That is the claim, and it is about the combination. On single rows, others even win, and the page says so.

• Ahrefs has less friction. Its checker needs no signup and has no limit. The Snapshot needs no account either.
• Athena HQ’s free tier gives you more raw answers, about 300 credits spent once.
• Among paid one-offs, Cruelx asks each question three times on six engines, for $19.99.
• One-off agency audits run from $1,500 to $25,000. They go deeper on questions, and you get to meet someone.

What no free scan we checked does is match the engines’ answers to your own pages and turn raw data into insights. It is also the part that tells you what to fix.

Why give it away

The Snapshot is a diagnostic. Its job is to settle whether AI visibility deserves your time and budget at all. I think the information you need to make that decision should be free.

Right now it mostly isn’t, and what is free is often wrong. LinkedIn feeds are full of people who have already cracked AI visibility. Comment “GEO” and they will DM you the playbook. Very little of it holds up. Confident advice, hardly any data, a lot of guessing. Some of it contradicts the rest, and all of it sounds certain.

The paid route isn’t much better for a first decision. Agencies charge $1,500 to $25,000 for a one-off audit, and GEO retainers run to thousands a month, before anyone has shown you whether you have a problem worth paying for. That is a lot of money to spend on finding out whether to spend money.

So the Snapshot gives you the data first. Your questions, your engines, your competitors, the sources behind every answer, and a report where every claim links back to what the engines actually said. You can check all of it. Then you decide, with evidence, whether to do something about it, whether to do it yourself, and whether to pay anyone at all, including us.

A price in front of that decision only filters for the people who have already made up their minds.

The Snapshot is free to run. Extra Snapshots are available as credits.

If you run an agency, you can buy Snapshot credits at an agency rate and run them for your prospects and clients. We are open to working with agencies that want to use the Snapshot to show their customers where they stand. Write to hello@heralded.ai.

What made it possible for free

When we launched in June, the full scan was a half-manual process. It asked four engines ten questions, and it cost €149, because a lot of the work behind it was still done by hand.

The next step was a small, light version we could afford to give away. Useful for a first look, and deliberately limited.

Today’s Snapshot asks six engines twice as many questions, twice each, reads the cited pages and your own site, and costs you nothing. Most of that change happened in the last month, and most of it comes from new technology we started using in September.

Asking the engines is the visible part. The expensive part is everything after it. A single Snapshot makes thousands of small judgements. Is this cited domain a review site, a competitor or a trade publication? Does this answer name the brand, or a product with a similar name? Is this rival a real substitute for the same buyer, or a company that happens to share a keyword? Which of the brand’s pages answers this question, if any?

Those judgements used to be the slow, costly part of every Snapshot. Now they are fast, more accurate and cheap. A Snapshot still costs us money to run, but little enough that we can cover it ourselves. And give one away for free to you.

Under the hood

For the curious, here is what changed.

Until September we made those calls with a mix of regular expressions, keyword lists and general chat models asked in batches. The chat models were slow and expensive. The rules were fast, cheap and confidently wrong in ways nobody notices until a report looks odd.

In September we moved most of them to Jev, the new System One model from TypeSafe. It answers narrow questions over structured input. Either it picks one of a set of named options, or it gives the probability that a yes-or-no statement is true. There is no essay to parse afterwards.

We didn’t switch on faith. Each decision was tested against the version running in production, on the same real records, and scored against references labelled by reviewers who never saw either system’s output. Those reviewers were AI agents rather than people, which is worth keeping in mind. A few of the results:

• Sorting cited sites into review platforms, media, competitors and the rest got about 40% more of them right, around forty times faster, at about half the cost.
• Wrongly admitted competitors fell by more than 80%.
• Matching a buyer’s question to the right page on your site cut wrong matches by more than half and missed matches by over 40%, for about a third less.
• Where the old rules and Jev disagreed about whether an answer named the brand, Jev was right every time and the rules about a third of the time. Many of the rules’ misses came from one brand with a curly apostrophe in its name. Typography remains undefeated.

The cost change was just as large. On one test domain, the bill for these judgements fell by more than 90%.

The full engineering write-up, including every decision Jev lost, comes separately.

After the Snapshot

A Snapshot is one reading on one day. It tells you where you stand. It can’t tell you whether anything you change moves the answers, because that takes the same questions asked every day, and somebody doing the work in between.

That is the subscription, which launches on 6 October. It reads the six engines daily on the questions you choose, and agents do the work the Snapshot points at: they write the pages you’re missing, fix your site through your CMS, and draft the pitches to the sites the engines cite. You approve what ships.

Run your free Snapshot at heralded.ai. About ten minutes. No account needed.