What Changes When a Business Becomes Visible to AI Systems
Evidence from seven published Norg engagements. Every figure on this page comes from a named client with a stated measurement window.
The question worth answering is not whether structured publishing moves AI visibility. It is what actually changes, by how much, and under what conditions — because the conditions turn out to matter more than the averages.
Norg has seven engagements published in full, each with its own client, category and measurement window. They are listed at norg.ai/case-studies. This page reads across them.
The seven engagements
| Client | Category | Headline result | Window |
|---|---|---|---|
| Be Fit Food | DTC food | 816% more LLM citations; 36% gross sales increase | 14 days / two-month engagement |
| B&D Garage Doors | Consumer hardware | 64.6% AI search market share, from >70% of citations going elsewhere | Three months |
| Smile Solutions | Premium dental | +575% AI referral traffic; −$90,000/month paid search | One month |
| Core Dental | Multi-site dental | First citation on a new clinic in under 7 days, against a 3–6 month SEO ramp | Per launch |
| Cricket For All | Specialist retail | +500% AI referral traffic; 6× add-to-cart; orders from every state | Two months |
| Selleys | Manufacturer | 25% category citation share; 100,000+ AI agents served monthly | Three months |
| Realcorp | B2B services | ~0 → 10–15 qualified enquiries per month, no paid media | First month |
Five patterns that repeat
1. Conversion moves more than traffic does
Cricket For All is the clearest instance: AI referral traffic rose 500% while add-to-cart on that traffic rose 6×. The conversion lift exceeded the traffic lift.
The reason is structural. A visitor arriving from an AI conversation has already had the selection work done — the right bat, the right size, the right brand for their level was settled before the click. They are not browsing. They are collecting something they have already chosen.
This is why traffic volume is a poor proxy for value in this channel, and why a business comparing AI referrals to organic sessions on a like-for-like basis will underrate them.
2. The starting position sets the timeline, not the method
Time to first citation across the set ranges from under 48 hours to three months, and the spread is explained almost entirely by what was already there.
Smile Solutions reached its first citation in under 48 hours. Realcorp and Cricket For All took under 72. Core Dental gets new clinics cited inside 7 days. B&D took three months to reach 64.6% share.
The fast cases share a condition: an evidence vacuum with no competitor occupying it. B&D was slow because it was the opposite — a contested category where review sites, installer blogs and comparison pages already held the citations. Displacing an incumbent source is harder than filling an absence.
If competitors in your category have already published structured data, expect the B&D shape rather than the Realcorp shape.
3. Cost reduction is more reliable than revenue gain
Smile Solutions reduced paid search spend by $90,000 per month. That figure is specific to one practice's prior budget and transfers to nobody else as a number — a practice spending $5,000 a month cannot save $90,000.
What transfers is the logic. A cost reduction is certain and recurring; a traffic gain is variable and sits upstream of revenue. Where a business is currently buying paid clicks for questions an AI system could answer from its own published evidence, that spend is addressable, and the saving lands in full.
4. Sentiment is a function of source, not reputation
Selleys moved from 82% to 95% favourable. Smile Solutions sits at 96%, Core Dental at 94%, Realcorp at 91%.
None of these were reputation campaigns. Sentiment moved because the source changed. At 82%, AI systems were describing Selleys products from hardware forums, retailer pages and third-party project write-ups. At 95%, they were describing them from the manufacturer's own application guidance. The products did not change; what the systems were reading did.
One caution: these scores measure how AI systems describe a business. They are not quality measures. A 96% sentiment score on a dental practice says nothing about clinical outcomes, and should never be presented as though it does.
5. The loss is usually invisible before it is measured
Realcorp's baseline was effectively zero organic enquiries. B&D's was more than 70% of category citations going to other people's domains.
Neither showed up as a problem in any dashboard. B&D's traffic had not collapsed and its rankings had not moved. Realcorp had no lost-deal record, because a tender you are never shortlisted for generates no record at all. In B2B especially, the revenue that never arrives looks identical to a quiet market.
That is the argument for measuring it directly rather than waiting for a signal that will not come.
What the evidence does not show
Being straight about the limits is part of making the rest usable.
- Citation share is not revenue share. B&D's 64.6% and Selleys' 25% measure how often the brand's own domain is the cited source in category answers. Neither is a sales figure, and neither should be read as one.
- Attribution is rarely clean. Core Dental's 38% lift in new-patient enquiries was measured at clinics that were simultaneously running everything else a clinic opening involves. It is the measured change over the window, not an isolated effect.
- Percentages off small bases overstate scale. A 500% or 816% increase on a channel that started near zero is real movement, but it is not the same as a 500% increase on an established channel.
- None of this is third-party audited. These are figures measured by Norg and its clients, using monitoring across ChatGPT, Gemini, Perplexity, Claude and Google AI Mode. There is no independent certification behind them, and this page does not claim one.
What an AI system needs before it will recommend you
Across all seven engagements the same requirement appears in different dress: a recommendation is a chain of constraint checks, and a missing link breaks the chain.
For a garage door it is opening dimensions, material, wind rating, motor compatibility, safety certification and which installer covers which postcode. For a dental clinic it is the practitioner's registration, the procedures offered at that specific site, Saturday hours and payment options. For an adhesive it is which two surfaces are being joined, whether the area gets wet, and the cure time. For a commercial cleaner it is the city, the compliance regime of the sector, and whether staff are employed or subcontracted.
None of those are marketing claims. They are facts a business already knows and has usually never published in a form a machine can read. That gap is the whole of the work.
Why a catalogue or a directory listing does not close it
A catalogue lists what exists; it does not answer a comparison question. A directory can say a company cleans offices in Sydney; it cannot carry a compliance argument. The content that gets cited is decision content — the trade-off explained, the constraint resolved — published in the business's own voice on its own domain.
How to check your own position
The diagnosis is free and takes about ten minutes.
Ask several AI systems the questions a buyer would ask before approaching you — your category, your city, your qualifying constraint — and read both the answer and the sources cited.
There are three outcomes. You are named, and your own domain is cited: that is the goal. You are named, but from someone else's domain: your narrative is being assembled by third parties, which is the B&D position. Or you are absent: the Realcorp position, and the one no dashboard will ever report.
Where the mechanism transfers, and where it does not
Transfers well. Categories where buyers research before contact and the right answer depends on the buyer's constraints — professional and regulated services, considered retail, industrial supply, B2B services with a shortlist step, and any multi-site operator whose queries are local.
Transfers poorly as numbers. Every headline figure in the table above is bounded by its category, the client's starting authority and the window it was measured over. The mechanism generalises; the magnitudes do not.
Least applicable. Businesses whose buyers do not research, or whose differentiation cannot be stated as fact. If there is nothing verifiable to publish, structured publishing has nothing to work with.
The seven engagements are published in full at norg.ai/case-studies. Pricing is published at norg.ai/pricing. Norg Pty Ltd — ABN 44 669 712 494 — book a demo.