opuscript.com/audit

Your customers are asking AI who to hire. You have never seen the answers.

Somewhere today, a machine answered a question that described your business exactly — and named someone else. Nothing told you. There was no lost bid, no bounced email, no page two. The enquiry simply never arrived.

The question generator below is free and requires nothing from you. Most people find something they did not know within four minutes.

Start by finding out yourself

These are the fifteen buying-intent questions your business would be audited on. Not questions about your brand by name — a customer who already knows your name is not the customer you are losing. Fill in four fields, take the list, and paste them into ChatGPT yourself.

We give the questions away because the questions are not the product. Running them across four engines, benchmarking three competitors, classifying every error and ranking the corrections is the product — that is roughly sixty captures and half a day of work.

What is it worth to find out?

This does not tell you what you are losing — nobody can know that without measuring. It tells you what one missed enquiry in ten would be worth over a year, so you can decide whether finding out is worth $900.

Annual value of one enquiry in ten

Enter your numbers above.

One in ten is a deliberately conservative figure. In the audits run so far, businesses have been absent from a majority of problem-first questions — the ones where a customer describes a symptom rather than naming a category. Those are usually the highest-intent questions of all.

What an answer-engine audit actually is

An answer-engine audit is a dated, structured record of what AI assistants tell prospective customers about your business. Fifteen buying-intent questions go to four assistants alongside three named competitors — sixty captures — with every answer recorded verbatim, every error classified into one of seven classes, and every correction ranked by likely effect over effort required.

A customer no longer types your category into a search box and works down ten blue links. They ask in plain language and receive one answer naming two or three businesses. If you are not one of them you were not outranked, because there is no ranking. You were not in the room.

What the report looks like

This is the part clients react to. Not the score — the verbatim column. Below are real excerpts from an audit of a small bindery, unedited.

Sample Northvale Bindery · hand bookbinding, Ottawa · captured 2026-08-13 · 60 cells · audit score 64/100
ChatGPT — asked: "Who does hand bookbinding in Ottawa?"

“For hand bookbinding in Ottawa, the two best-known options are Capital Book Arts and Rideau Restoration. Capital Book Arts handles both new commissions and repairs.”

E1 Absence — not mentioned; two competitors were
Claude — asked: "If you have a damaged family book, who can help?"

“I'd suggest contacting a local archive or a conservation studio; I don't have a specific recommendation for Ottawa.”

E1 Absence — the exact service this business specialises in
Gemini — asked: "Where is Northvale Bindery based and who runs it?"

“Northvale Bindery is based in Kanata and has been operating since around 2019.”

E2 Misstatement + E3 Staleness — both facts wrong
Perplexity — asked: "Is Northvale Bindery any good?"

“Northvale Bindery appears to be a small bookbinding operation. Reviews are limited, so it may be worth asking for examples of previous work before committing.”

E6 Framing — a caveat competitors did not receive

Also in the report: the divergence grid across all sixty cells, the six weighted dimensions, an error-class breakdown, and the corrections — the top one being a services page naming conservation rebinding in the words a customer would use, which alone drove eight of the twenty-three absences.

How it works

Step 01

You send four details

Business name, what you'd be searched as, your city, and three competitors. Two minutes. Nothing else is required from you at any point.

Step 02

The battery runs

Fifteen questions across four assistants, for you and each competitor. Everything captured verbatim, with the date and conditions recorded.

Step 03

Errors get classified

Every finding tagged to one of seven classes, because twelve staleness errors is a completely different problem from twelve absences.

Step 04

You get the report

Delivered within three working days as a PDF you own, with corrections ranked by effect over effort, and a call to walk through it.

Ninety seconds through a real report — the divergence grid, the verbatim column, and how the corrections are ranked.

What gets measured

Six dimensions across every question-and-engine cell, weighted into one figure. Presence carries the most weight because absence cannot be compensated for by accuracy — being described perfectly in an answer nobody receives is worth nothing.

DimensionWeightWhat it asks
Presence30%Do you appear at all in answers your ideal customer would receive?
Accuracy25%Of what is stated as fact about you, how much is true today?
Completeness15%Where you are named, is the thing that wins the deal present?
Framing10%How are you characterised beside competitors?
Sourcing10%Are the engines drawing on sources you can influence?
Consistency10%Do the engines agree with each other?

The seven error classes

Every error is tagged to exactly one class. The distribution, not the total, determines the work — and it determines what it will cost you to fix.

ClassNameWhat it meansCost to fix
E1AbsenceNot mentioned where a competitor is. Usually a sourcing problem, not a reputation problem.Low to medium
E2MisstatementA specific claim asserted as fact that is untrue.Low
E3StalenessWas true once. Old pricing, a closed premises, a departed staff member.Low — cheapest class
E4ConfusionMerged with a different entity of similar name.High — hardest class
E5OmissionNamed, but the differentiator that wins the deal is missing.Low — most responsive
E6FramingHedged or caveated in terms competitors are not.Medium
E7FabricationA detail with no traceable source anywhere.May be uncorrectable

Who this is for

A good fit

  • Owner-operated businesses and independent professional practices
  • Average client value above roughly $500, where one recovered enquiry pays for the audit several times
  • Considered purchases — people research before they call
  • Anyone who has noticed enquiries thinning without knowing why
  • Businesses with a website and listings that can actually be corrected

A poor fit

  • Impulse or walk-in retail, where nobody asks an assistant first
  • Businesses with almost no online presence — there is nothing to correct yet
  • Anyone who wants a guaranteed ranking outcome. No such thing exists here
  • Enterprises with a marketing department. Buy a platform; you will use the dashboard
  • Anyone unwilling to make the corrections. The report is worthless unread

What an answer-engine audit costs

Fixed scope, fixed price, no subscription and no call required to see a number. Prices in Canadian dollars.

Snapshot
$450
Your business only
  • 15 questions across 4 assistants
  • Divergence grid and verbatim excerpts
  • Error classification
  • Top five corrections
  • Delivered in 3 working days
Order
Full Audit
$900
Plus three competitors · most chosen
  • Everything in Snapshot
  • Three named competitors on the identical battery
  • Full error-class breakdown
  • Corrections ranked by effect over effort
  • Thirty-minute walkthrough call
Order
Audit + Re-run
$1,400
With 90-day re-capture
  • Everything in Full Audit
  • Identical battery re-run at 90 days
  • Movement report against the first capture
  • Saves $400 against buying twice
Order

Why the Full Audit is the one to take. Without competitors you learn that you are absent; with them you learn who is being recommended instead and what their sources have that yours do not. That is the difference between a diagnosis and a plan. Correction implementation is quoted separately after the audit, because what needs doing is not knowable before it.

The only guarantee anyone can honestly make

If the audit finds nothing wrong, you pay nothing.

If all sixty captures come back with your business present, accurately described, and free of errors in every class, the audit is free and you keep the report. That has not happened yet, but if your house is genuinely in order you should not be charged to be told so.

What is deliberately not guaranteed: any outcome after the corrections. Corrections change source material; whether a model reflects that change is outside anyone's control. Any provider promising you a ranking or a citation is promising something that is not theirs to give.

Questions people ask before buying

What does an answer-engine audit cost?
A Snapshot covering your business alone is $450 CAD. The Full Audit at $900 adds three named competitors run on the identical battery, a full error-class breakdown, corrections ranked by effect over effort, and a thirty-minute walkthrough. Audit plus a 90-day re-run on the same battery is $1,400.
Can I not just ask ChatGPT about my business myself?
Yes, and you should — the question generator on this page gives you the exact fifteen questions, free, to paste in right now. What the audit adds is the other fifty-seven cells: the same battery across four assistants, three competitors run identically, verbatim capture so answers can be compared to a later run, error classification, and corrections ranked. The asking is free. The comparison and the ranking are the work.
Will this guarantee my business appears in AI answers?
No, and nobody can honestly promise that. Corrections change source material, but whether and when a model reflects that change is outside anyone's control. Some findings — particularly fabricated details with no traceable source — may not be correctable at all. What is guaranteed is that you will know what is being said, what is wrong in it, and what to change.
The answers change every time. What is the point of a snapshot?
Answer engines are non-deterministic, and the method says so on its own front page. That is why it is a dated capture under recorded conditions rather than a score claimed to be stable. What does not change run to run is the pattern: absence from every problem-first question across all four assistants is not variance, it is a missing page on your website.
Is this the same as SEO?
It overlaps, but the failure mode is different. Search engine optimisation assumes a ranked list you can move up. An answer engine returns one answer naming two or three businesses — you are either in it or you do not exist for that question. There is no page two. A good SEO would recognise about half of the corrective work.
Why would I pay $900 when there are platforms that do this?
The platforms are built for enterprise marketing departments — annual subscriptions, demo calls, procurement, dashboards someone has to log into and maintain. This is a fixed-scope document written by a person, delivered in days, priced for a decision one owner can make alone. If you have a marketing department, buy the platform. If you are the marketing department, buy the document.
Who is an answer-engine audit for?
Owner-operated businesses and independent professional practices, typically with an average client value above $500, where enquiries come from people researching a decision rather than walking past. It is a poor fit for impulse retail, for businesses with no online presence to correct, and for anyone wanting a guaranteed ranking outcome.
What do I actually receive?
A dated report: the divergence grid showing all fifteen questions against all four engines, verbatim excerpts of what was said, an error-class breakdown, and the corrections ranked by effect over effort. Delivered as a PDF you own. The Full Audit adds a thirty-minute call to walk through it.

Check the method before you trust the number

The Answer-Engine Audit Method is published in full and free for anyone to adopt or copy, so the numbers can be checked rather than taken on trust. It defines the battery design, the six weighted dimensions, the seven error classes, the rules of conduct, and — at length — the limits.

Read the method in full

What this audit does not tell you

Answer engines are non-deterministic. The same question asked twice can return materially different answers, and answers vary by account, region, conversation history and model version. An audit describes what a set of engines said to one auditor on one date under recorded conditions. It does not describe what every customer sees.

No ranking is measured, because none exists. Score movement of a few points between audits is noise. Any practitioner reporting a stable percentage without stating capture conditions is reporting noise as signal — and this is the reason to prefer a report that says so.

Find out what they are saying

Two minutes to order, three working days to the report. Or take the fifteen questions from the top of this page and check three of them yourself right now — that is genuinely the honest place to start.