Herald · publish and keep score
An AI visibility tool that keeps the receipt.
When someone asks an assistant a question in your market, either your business gets named or it does not. Herald asks those questions on a schedule, records who was named, and keeps the answer it came from.
The log
What the log looks like.
Every entry keeps the question, the assistant, the date, and whether you were named. The rows below show the format. They are an example, not anyone's real data.
| Date | Question asked | Assistant | Result |
|---|---|---|---|
| 14 Jul 2026 | Who supplies bulk cleaning products in Brisbane | Assistant A | Named |
| 14 Jul 2026 | Best commercial cleaning supplier for offices | Assistant A | Not named |
| 13 Jul 2026 | Where to buy eco friendly cleaning supplies in bulk | Assistant B | Named |
| 13 Jul 2026 | Cheapest bulk hand sanitiser suppliers | Assistant B | Not named |
The useful column is the last one, read downwards. A list of the questions where you never appear is the only part of this you can act on.
The method
How the tracking works.
1. The questions
Scout builds the list from what people actually search for in your market, not from terms you guess at. You can add your own.
2. The checks
Herald asks each question on a schedule across the assistants being tracked, and records the full answer alongside the result.
3. The gap list
Questions where you are never named become work: Scout queues them, Scribe drafts, and Checker verifies before anything publishes.
Definition
What an AI visibility tool actually does.
An AI visibility tool checks whether AI assistants name your business when someone asks a question in your market. It runs a fixed list of questions across the assistants on a schedule, records which businesses were named and which pages were cited, and tracks how that changes over time.
That is the whole category. The differences between products are in how many questions they run, how many assistants they reach, how often they run, and whether you can see the evidence behind the number they show you.
It is worth being clear about what this is not. It is not a ranking report, because assistants do not publish rankings. It is not a guaranteed placement, because there is nothing to buy. And it is not a measurement in the strict sense, for the reason set out further down this page.
Coverage
Which assistants Herald checks, and which it does not.
Herald samples the assistants that publish an interface we are allowed to call:
- ChatGPT
- Claude
- Gemini
- Perplexity
Three surfaces that matter are missing from that list, and we would rather name them than let you assume: Google AI Overviews, Google AI Mode and Microsoft Copilot. None of them publishes an interface for this. The practical way to report on them today is to drive the consumer product as though a person were typing into it, which is contrary to those products' terms of use.
Some tools in this category do report on those surfaces. We are not going to tell you how they do it. We are telling you what we do, which is that we do not, and that a number we cannot obtain honestly is one we would rather leave blank.
That is a real limitation and it cuts against us commercially, because the Google surfaces appear in front of people who never chose to talk to an assistant. What the four we do sample give you is a genuine reading of the assistants people deliberately ask, taken the same way every time.
The list will change as these products publish interfaces or withdraw them. Your dashboard shows which surfaces ran on any given check, which is the version that is always current.
One more thing this industry leaves out. Assistants rate-limit, change without notice and occasionally fail outright. When a run cannot reach an engine, that run is recorded as an error and reported as one, with the sample count and the error rate next to every figure. A tool that silently drops failed samples shows you a cleaner number than it earned, and you cannot tell a quiet week from a broken connection. If an engine went dark for a fortnight, you should be able to see that it did.
Two things, not one
Mentions and citations are not the same thing.
These get merged into a single score by most tools, and they should not be.
A mention is the assistant naming your business in the body of its answer. The reader sees your name. They may never visit your site, and there may be no link at all. A citation is the assistant pointing at a specific page as a source, which is the part that can send someone to you.
You can have either without the other. Being named without being cited is common, and it still has value, in the way being recommended by name in a conversation has value. Being cited without being named happens too, usually when a page of yours is useful and your business is not the subject of the question.
Herald records them separately, along with the question asked, the surface it was asked on, the date, and the answer text. A number you cannot trace back to the reply that produced it is not evidence, it is a claim.
The field
How this compares with the dedicated trackers.
There is now a real category of AI visibility products, and several of them are good. If pure tracking is what you want, they are worth looking at: Profound, Peec AI, Otterly, Rankscale and Ahrefs Brand Radar are the names that come up most, and the larger SEO suites have added trackers of their own.
Two things are worth knowing before you compare them. The first is that engine coverage is usually the upsell: entry plans in this category are frequently limited to one assistant, with the others sold as add-ons or reserved for higher tiers. Check what the cheapest plan actually reaches before comparing headline prices.
The second is that a good number of them do not publish prices at all. Several list tiers with feature rows and no figures, and route you to a sales call instead.
The honest difference here is not that Herald tracks better. It is that tracking is not the product. In a dedicated tracker, the output is a dashboard and the work of responding to it is yours. Here the gap list is an input: questions you are never named in become drafts, which pass the standard or are rejected with reasons, and publish if they pass. If you want a tracker, buy a tracker. This is for people who want the follow-up work to happen.
Cost
What it costs.
Citation tracking is not sold separately. It is part of every Searchlings plan, because a product that publishes content and cannot tell you whether it is being quoted is only doing half a job.
- Starter, $299 a month: 30 tracked questions, checked monthly.
- Growth, $699 a month: 100 tracked questions, checked weekly.
- Scale, $1499 a month: 250 tracked questions, checked daily.
Prices are in Australian dollars, including GST. Full inclusions are on the pricing page. A first check costs nothing at all: send a Scout and it will tell you whether assistants name you today.
What moves it
What actually earns a citation.
Tracking tells you where you stand. It does not tell you what to do, and a dashboard that only reports is a bill you pay to feel informed. Here is what the available evidence suggests actually changes the answer.
- Answer the question in the opening paragraph. Models lift openers. CXL sampled 100 AI Overview citations and found 55 per cent came from the first 30 per cent of the source page. Burying the point below three paragraphs of preamble is the most common and most fixable error.
- Write sentences that survive being quoted alone. A sentence beginning "as mentioned above" cannot be extracted. It only means anything in place.
- Keep your facts identical everywhere. If your services, location or founding year differ between your site, your LinkedIn page and a directory listing, a model has no principled way to choose. The usual outcome is that it avoids the claim entirely.
- Have something specific to say. Models summarise consensus. A page that restates the consensus is absorbed into it without attribution. A number you measured or a method you use is the thing that gets quoted.
- Let the crawlers in. A page that cannot be fetched cannot be cited, and this is a surprisingly common own goal.
Ranking well is not the same as being quoted, and the gap is widening. Ahrefs studied 863,000 keyword SERPs in March 2026 and found 38 per cent of AI Overview citations came from pages ranking in the top 10, down from roughly 76 per cent in July 2025, with about 31 per cent coming from pages ranking beyond position 100. A rank report is a weakening proxy for whether you are being quoted.
The longer versions of this argument are in the Answer Engine Optimization article and the Generative Engine Optimization article.
The limits
The honest limits of any AI visibility tool.
Assistants do not publish rankings, do not guarantee consistent answers, and do not explain why they named one business over another. Any tool claiming a precise share-of-voice percentage for AI answers is presenting a sample as a census.
What tracking honestly gives you is a direction of travel over weeks, and a list of questions you are absent from. That is genuinely useful. It is also all of it, and we would rather say so here than in a support ticket.
Questions
Questions.
Which assistants are checked?
The assistants that publish an interface we are allowed to call: ChatGPT, Claude, Gemini and Perplexity. Google AI Overviews, Google AI Mode and Microsoft Copilot do not publish one, and reaching them means driving the consumer app against its terms of use, so we do not. Your dashboard shows which surfaces ran on any given check.
Why do results change between checks?
Assistants do not return the same answer every time. The same question can name different businesses on different days, from different locations, in different sessions. This is sampling, not measurement, which is why the trend over weeks is the number worth watching.
Can I pay to be cited?
No. There is no placement to buy in an assistant answer today. Citations come from being a source the model has reason to trust, which is slower and considerably harder to fake.
How long does it take to earn a citation?
Longer than a ranking change. Models are retrained and their indexes refresh on schedules nobody outside those companies controls. We do not publish a typical timeline, because we would be inventing it.
What is the difference between a mention and a citation?
A mention is the assistant naming your business in the text of its answer. A citation is the assistant linking to a page as a source. You can have either without the other, and they are worth different things: a mention reaches the reader, a citation sends them to you. Herald records both separately rather than merging them into one score.
Do AI visibility tools fix anything, or only report?
Most only report. That is a legitimate product, and if you have a team to act on it, it may be all you need. Here the gap list feeds the work directly: questions you are never named in become drafts, which are checked against the standard and published if they pass.
What is a good AI visibility score?
We do not publish one, because a single blended score across assistants that behave differently would be a number with no meaning. The comparisons worth making are against your own previous weeks and against the specific questions you care about.
How many questions are checked, and how often?
It depends on the plan. Starter tracks 30 questions monthly, Growth tracks 100 weekly, and Scale tracks 250 daily. Every run records the answer text it came from.
Is there a free way to check?
Yes. Send a Scout runs a first check on whether assistants name your business, with no account and no payment. It is a single sample rather than a tracked series, so treat it as a starting reading rather than a trend.
Find out if assistants mention you.
Give us a web address. Scout will check whether AI assistants name your business when people ask, along with the search basics and your site health. Free, no account needed.
Send a Scout