Should I hire an agency or buy an AI visibility tool?

Are they even alternatives?

Not really, and that is the part of the question worth correcting first.

A tool and an agency do not compete for the same job. One produces a reading. The other changes the thing being read. Comparing their monthly prices is like comparing a thermometer with a plumber: the thermometer is cheaper every month of the year, and it will never fix the hot water.

The useful version of the question is do I need a number, or do I need the number to move? Almost everybody needs the first before they can sensibly buy the second, and plenty of businesses should stop after it. There is a question underneath both, which is whether any of this is separate work at all, and that one is answered on is AI visibility just SEO with a new name.

What does a monitoring tool actually do well?

Four things, and it does all four more cheaply and more consistently than a person would.

• It repeats. The same questions, on a schedule, without anyone remembering to run them. Repetition is the whole value, because a single reading of a system this noisy is close to meaningless.
• It keeps a record. A dated series you can point at later is worth more than a better one-off report, and it is the only way to tell a real move from a model update.
• It catches change you would not notice. Nobody reads their own category's AI answers every week. Software does.
• It keeps whoever does the work honest, including an agency. If one supplier owns both the work and the scoreboard, you have no independent check on either. A separate measurement is a governance control, not just a report.

None of that requires a person, and paying agency rates for it is a poor trade.

What can a tool not do?

Change the answer.

Every gap a scanner reports is closed by somebody making an edit somewhere. In our 30 July 2026 scan of 300 Australian businesses, 13% of the 300 blocked AI crawlers outright in robots.txt and 24% of the 300 had no JSON-LD structured data on the homepage. A tool can tell you that in a second. Somebody still has to open the file, decide whether the block was deliberate, and change it. That decision is not always obvious either, which is the point of whether you should block AI crawlers at all.

The harder limit is judgement. Across the 300 businesses in that scan, on-site readiness averaged 74.9 out of 100, yet across the 137 with complete records from both engines it correlated with per-engine visibility at r = 0.116. That is close to nothing. So a tool that hands you a tidy list of on-site fixes is handing you the part of the problem our own data says barely predicts the outcome. Working out which gap is costing you the mention is analysis, and analysis is a person.

When is a tool on its own the right call?

When somebody is already in place who can act on what it says.

• You have an in-house marketer, a developer or a capable agency already doing your website and content work, and what is missing is the scoreboard.
• You already know your gaps and want to watch whether closing them does anything.
• You hold a position you are happy with and want an alarm if it slips.
• Your budget is small enough that you would rather spend it on the work than on somebody describing the work.
• You are testing whether this category matters in your industry before committing to anything.

In all five cases, buy the cheapest thing that asks real buyer questions of real engines and shows you the answer text. If it will not show you the answers, it is not evidence.

When is a tool on its own the wrong call?

When nobody is going to do anything with the reading.

Monitoring bought before a diagnosis is a subscription to a graph. If you are one of the businesses never named at all, and 41% of those 137 complete two-engine records were never named once by either engine across every question asked, then your monthly chart is a flat line at zero. Twelve of those in a row is not information, it is an expense.

The opposite case is just as bad. 26% of the same 137 were recognised by the engines when asked about directly, yet never named when a buyer asked who to use. A score on its own cannot separate those two situations, and they need completely different work. That distinction is what being known but never recommended is about.

What is the honest cost comparison?

Tools are cheap, recurring and predictable. Implementation is lumpy, mostly labour, and priced per situation.

What nobody can give you is the figure that would actually settle the decision: a credible cost per additional AI mention, or a conversion rate from a mention to a customer. Nobody has measured that, including us. There is no public dataset linking AI answer mentions to Australian revenue, and any vendor quoting you one has invented it.

Use a method instead of a number, and the long version of that method is on how much a ChatGPT mention is worth. Work out what one new customer is worth to you over a year. Divide the annual cost of whichever option you are weighing by that figure. That gives you the number of customers the spend has to produce to break even, which is something you can reason about from your own close rate. If the break-even is one or two customers a year, the decision is easy. If it is thirty, it is not.

We sell both, so here is the disclosure

Answerable sells the measurement and the implementation, which means we have a commercial interest in whichever way you answer this.

Our ladder is a free scan, a $495 one-off audit, monitoring at $199 a month that we will not sell before an audit or a fix, and custom-scoped done-for-you work. Monitoring is gated behind a diagnosis for exactly the failure mode above: a subscription to a flat line is not a product. Prices and the comparison against published tool pricing are on the cost page, and the scoring formula behind any number we give you is published in full on The Receipts Engine.

The argument against buying a large package from anyone, us included, is in our own dataset: r = 0.116 between site readiness and per-engine visibility, across those 137 businesses. If a supplier sells you a bundle of on-page fixes as though it were the whole job, that number is the thing to quote back at them.

How do I decide, in one pass?

Answer four questions in order and stop at the first no.

1. Do I know my current position? If not, get a reading before buying anything at all. A free scan is enough to start, and you can do a rough version by hand.
2. Is the answer actually wrong about me? If the engines already name you for your real buyer questions, buy monitoring and stop. There is nothing to fix yet.
3. Do I have hands that can fix it? If yes, buy the tool and brief your own people. If no, this is where paying somebody starts to make sense.
4. Will anything be re-measured on the same questions? If not, buy neither. Unmeasured work in this category is indistinguishable from no work.

Start with the reading, whichever way you go.

The free scan gives you a score, the questions we asked and who was named instead of you. It is the input to this decision, not a commitment to either side of it.