AI Visibility Audit for marketplaces · AI Discovery Intelligence
Ask ChatGPT, Claude, Gemini or Perplexity where to buy or sell in your category and it names a destination. That selection happens before any search, off your property, where your analytics never see it. If you are not the pick, nothing downstream fires - not the session, not the lead, not the renewal.
We measure your share of AI recommendations on the queries that carry revenue, and we do the work that moves it. This is a selection problem, not a ranking problem.
200+ marketplaces measured · 15 countries · 4 engines · every read confidence-tagged
Measured from the outside. No SDK in your rendering path. We do not measure training-data exposure - and we say so.
What we have measured
to date
Every figure re-derives from the answer record. We publish what we measured, the denominator it sits on, and what we cannot see.
What the data actually says
Across every marketplace we have measured, branded recall is near-perfect - ask an engine about the company by name and it answers well, every time. The gap opens on the unbranded question, where the buyer has not decided yet and the engine picks for them.
The measured pattern is consistent: an engine names the marketplace as a place to look, then recommends a specific provider. Being the reference is not the same as being the pick, and the two move independently. We measure both separately.
When a buyer question names two rivals, marketplaces appear as an aside rather than the answer - and in the sets we have measured, almost never as the pick. Engines also tell buyers the portals carry the same supply from the same sellers, which is the argument you have to beat.
Financing, mortgages, valuation, the paperwork - the highest-intent moments route to banks, brokers and app stores rather than the marketplace. It is the part of the funnel you least want decided elsewhere, and the easiest to lose without seeing it.
Patterns above are drawn from marketplace runs across auto, property, and digital-goods categories in multiple markets. We report the pattern and its basis; we never expose another marketplace's numbers to make a point.
The board-level risk
A marketplace's moat is being the place buyers and sellers go to discover, compare, and transact. Answer engines now sit in front of all three. The exposure is specific, and it is measurable.
Buyers ask the engine what is available and what it costs. It answers from whatever sources it trusts - which may not be your listings - so the discovery moment that used to start on your marketplace now starts inside a chat window.
When the engine answers the pricing, availability, and trust questions directly, the buyer has less reason to land on the marketplace at all. Your role as the place people go to decide is exactly what an answer engine is built to absorb.
The highest-margin moments - premium placement, lead handoff, monetisable intent - are where a competitor or a partner can get named ahead of you. That is the part of the funnel you least want decided off-platform, and the easiest to leak without seeing it.
The organic demand the engines used to send you now lands on whoever they recommend. To hold lead volume flat you rebuy it as paid - so a share loss you cannot see surfaces as a CPL rise you can.
Every figure we put against these is anonymised and confidence-tagged. We show the pattern and the magnitude; we never expose another marketplace's numbers to make the point.
The query universe
A marketplace's buyers do not ask four questions. They ask thousands, across a journey. The work is not to track 50,000 random queries. It is to curate the high-intent set by commercial intent and monitor the ones that actually drive revenue, mapped to where the buyer is in the decision.
High-intent queries near purchase, valuation, and selling carry most of the revenue. Open-ended research queries like "how the paperwork works" carry almost none. We weight what you track to where the money is - so the share number you read is the one that matters to the business, not a vanity average across the long tail.
What we find
These are patterns verified across multiple marketplaces in multiple markets, not category theory. Each one is a place a marketplace loses the pick without ever seeing it.
How often you are named tells you little. How often you are the pick once named separates pure aggregators from portals from single-brand operators, and it does so consistently across every qualified run we have measured. Visibility is the vanity number; pick rate is the commercial one.
Inventory, model and "where do people buy" questions go to the marketplace. Local, trust, service and by-name questions go to the seller - often through their profile page on your own platform. That routing decides which of you gets credited, and it is measurable per question.
Faceted pages get cited several times more often than seller profile pages, and name the seller far less often when they do. One template change fixes it across every seller at once - and no individual seller can fix it for you.
In the markets we have checked, most marketplaces block or challenge AI crawlers at the edge or in robots.txt - usually without anyone having decided to. Whoever opens selectively and deliberately first compounds the advantage, because the engines have nowhere else to get category truth.
Every finding carries its basis and its confidence tier. Where a pattern is strong but seen in fewer markets, we label it as emerging rather than established - and we tell you which is which before you act on it.
How we keep it honest
No one can prove a single AI recommendation caused a single lead. Anyone who claims they can is selling certainty that does not exist. We treat it the way marketing-mix modeling treats a channel: we estimate the contribution from converging signals and we label the confidence of every read, so your analysts can audit any claim back to the evidence behind it.
When a read is directly observed in your own data, we say so. When it is inferred, we say so. When it rests on industry intelligence, we say that too. You never get a confident-sounding number with nothing under it.
Directly observed. The signal is present in your own logs, CDP, or AI-engine output. We saw it happen.
Inferred. Multiple independent signals converge on the same read, but no single source confirms it outright.
Industry intelligence. The read rests on category benchmarks and external patterns, not your own data. Treated as directional.
In your stack
This is not a deck that ages the day it ships. It is a live read wired into the systems you already run, so the share number and its business impact stay current as the engines re-crawl and your competitors move.
The integrated engagement connects to your logs, CDP, and paid feeds, so the AI-share read sits next to the traffic and CPL data your teams already trust. No SDK in your rendering path.
You get a flag when your share on a high-intent query set moves sharply or a competitor surges, and a monthly read that tracks the trend and ties it to traffic and CPL. Not once a quarter, on a slide.
HIGH, MED, or LOW on every finding, so the team acting on it knows exactly how much weight it carries before committing engineering or budget against it.
Honest about the tiers. The live read on how much of your traffic and CPL movement is AI needs your first-party data - your logs, CDP, and paid feeds. That is the integrated, enterprise engagement, not a free snapshot. A standalone audit measures your share of AI recommendations across the four engines from the outside and shows where you are losing ground and to whom. The traffic-and-CPL loop comes when we wire into your systems.
What winning takes
A marketplace wins the pick when it becomes the source engines rely on to answer the category. That is a data-distribution problem, not a content-volume one. More posts will not do it.
The numbers buyers ask for before they decide. You hold transaction and pricing data engines currently fetch from third parties - which means someone else is authoring your category's truth.
What is actually listed, where, and at what price - reachable and structured so an engine can use it rather than guess around it.
The trust questions engines lean on when a buyer asks who to deal with. Today they often answer these from forums and third-party posts rather than from you.
Your most-cited pages are usually not the ones you optimised. We find them in the answer record and change them - template work, entity fixes, the structured facts engines are missing.
Around a dozen ranked moves per engagement, each one tagged so you know who does what: we deploy, we draft and you publish, or together. Precision over volume - the moves that shift measured answers, never a content quota. See what an engagement costs.
FAQ
An AI visibility audit measures how often an AI engine recommends your marketplace, versus a competitor, when a buyer asks where to buy or sell in your category. Built on AI Discovery Intelligence, 28 Labs scores a marketplace's share of AI recommendations across ChatGPT, Claude, Gemini, and Perplexity for the high-intent queries that drive revenue, then ties that share to real traffic and cost-per-lead movement and hands back the build moves that grow it.
Not thousands. We compose a panel of high-intent buying questions with you, you approve every one, and then it locks for the whole engagement so month two is comparable to month one. Tracking a huge rotating set is what makes a number move on its own; a locked panel is the only way a before-and-after means anything.
No one can prove a single citation caused a single lead, and we do not claim it. We model the contribution the way marketing-mix modeling does, and we disclose the confidence of every read: HIGH when directly observed in your own data, MEDIUM when inferred from converging signals, LOW when it rests on industry intelligence. We never present a false precision.
SEO optimizes ranked links on a results page. An AI visibility audit, built on AI Discovery Intelligence, measures and grows the recommendation an AI engine gives when a buyer asks it directly. Different surface, different signal. Winning it is a data-distribution problem - becoming the source AI trusts for pricing, inventory and availability, category benchmarks, seller reputation, and the category facts buyers weigh before they decide - not a content-marketing one.
Start here
A 60-minute working session. We walk a sample audit, show you how the share read and the traffic-and-CPL loop would shape up for your category, and scope a live run.
No obligation. We do not measure training-data exposure, and we tell you what we can and cannot see before you commit.