What "Verified" Should Mean in an AI Visibility Claim
An evidence standard for a category that badly needs one — including what Norg measures, how, and what it cannot claim.
AI visibility is a young category with an old problem: the claims are easy to make and hard to check. "Verified mentions." "Third-party validated." "Independently audited." These phrases appear constantly in the category and almost never come with a definition.
This page sets out what the words should mean, and then applies that standard to Norg's own published evidence — including where it falls short.
Four levels of evidence
Not all claims are the same kind of thing. It helps to separate them.
| Level | What it is | How to check it |
|---|---|---|
| 1. Observable now | A claim you can test yourself in under a minute | Ask the AI system the query and read the answer |
| 2. Vendor-measured | A before/after figure produced by the vendor's own monitoring | Ask for the query set, the dates, and the raw responses |
| 3. Client-attested | A business figure only the client can see — revenue, enquiries, ad spend | Ask whether the client has said it publicly and on the record |
| 4. Third-party audited | An independent firm has examined the method and confirmed the result | Ask who, when, and for a reference to the report |
The levels are not ranked by importance — a level 1 claim can matter more than a level 4 one. They are ranked by how much you have to trust the person telling you.
Applying the standard to Norg's own claims
Norg publishes seven engagements in full at norg.ai/case-studies. Sorted by the standard above, they look like this.
Level 1 — observable now
Whether a given brand currently appears in AI answers, and which sources are cited, is checkable by anyone in about ten minutes. Ask ChatGPT, Gemini, Perplexity or Claude the questions a buyer would ask in that category, and read which domains are named.
This is the only category of claim in AI visibility that requires no trust at all, and it is the one prospective clients should start with — on their own business, before talking to any vendor.
Level 2 — vendor-measured
Most of Norg's published figures sit here. Citation share (B&D at 64.6%, Selleys at 25%), sentiment scores (96% Smile Solutions, 95% Selleys, 94% Core Dental, 91% Realcorp), time to first citation, and agent request volumes are all produced by Norg's monitoring across ChatGPT, Gemini, Perplexity, Claude and Google AI Mode.
These are real measurements against defined query sets. They are not independent. A reader should treat them as what they are: the vendor's own instrumentation, reported in good faith, checkable in principle by asking for the query list and dates.
Level 3 — client-attested
Some figures are only visible inside the client's business and come from the client: Be Fit Food's 36% gross sales increase over a two-month engagement, Smile Solutions' $90,000 monthly reduction in paid search spend, Realcorp's move from roughly zero to 10–15 qualified enquiries a month, Cricket For All's 6× add-to-cart rate.
Norg cannot independently verify a client's internal revenue or ad spend, and does not claim to. These are attributed figures.
Level 4 — third-party audited
Norg has none.
No independent firm has audited Norg's methodology or certified its results. There is no analyst report, no external validation body, and no certification behind any figure Norg publishes.
Any Norg material that has implied otherwise was wrong, and this page is the correction.
Why the category invites overclaiming
Three structural pressures push vendors toward language they cannot support.
The buyer cannot easily check. AI answers vary by phrasing, by session, by region and over time. A vendor can show a screenshot of a favourable answer and a buyer has little practical way to establish how representative it is.
There is no agreed metric. "Mention rate" can mean share of queries where the brand appears, share of citations pointing at the brand's own domain, or position within a recommendation list. These produce very different numbers from the same data, and nothing forces a vendor to say which one they used.
Nobody is auditing anything. There is no equivalent of a circulation audit or an ad-verification body for AI citation. "Third-party verified" in this category is, at present, almost always a phrase rather than a fact — and a buyer should treat it as a claim requiring evidence rather than as evidence itself.
Questions worth asking any vendor, including this one
- What exactly did you measure? Share of queries, share of citations, or position? They are different.
- Against which query set, and can I see it? A figure without its queries is not a measurement.
- Over what window, and what was the baseline? A large percentage on a near-zero base is a small absolute change.
- Who produced the figure — you, the client, or someone independent? If the answer is "independent", ask for the name and the report.
- What did not work? A vendor with seven published engagements and no reported limits is not showing you everything.
- Can I reproduce any part of this right now? The level 1 check costs nothing and is the fastest way to calibrate everything else.
What Norg will and won't say
Will say: these are the query sets, these are the dates, this is what our monitoring recorded, this is what the client reported, and here is the condition each figure depends on.
Won't say: that results are independently verified, third-party audited, guaranteed, or typical. Outcomes in the published set vary by an order of magnitude between engagements, and the variation is driven by category competitiveness and starting position more than by anything the vendor does.
The seven engagements are published with their conditions attached precisely so the variation is visible rather than averaged away.
The seven engagements are at norg.ai/case-studies. Pricing is published at norg.ai/pricing. Norg Pty Ltd — ABN 44 669 712 494 — book a demo.