Getting found is now a retrieval problem.
Getting bought is now a data problem.
Visibility in AI assistants, discovery in search, and the product data that decides
whether a machine can transact with you at all.
the shift
The click economy is shrinking
faster than the AI channel is growing.
Both halves of that sentence matter. Most searches now end without a click, and the shift has accelerated sharply. Organic search accounts for a smaller share of web traffic than at any point in living memory. Where an AI summary appears above the results, the links beneath it get a fraction of the attention they used to.
Meanwhile the traffic arriving from AI assistants is still a sliver. But the people in it convert at or near the level of the best channel a brand has, they spend noticeably longer, and the volume is compounding fast. Not long ago this traffic converted worse than everything else. Now, in retail, it converts better.
So the honest framing isn’t “AI traffic is big.” It’s this: the channel you were built for is contracting, the channel replacing it is small and unusually high-intent, and becoming eligible for it takes months to compound. It’s infrastructure work. Start before you need it.
We’re deliberately not selling you agent-checkout readiness as a headline. The most prominent attempt at buying inside a chat assistant was withdrawn within months of launch — shoppers were researching there, then leaving to buy somewhere they trusted, and retailers measured materially worse conversion inside the assistant than outside it.
The behaviour that is real is comparison. Most people will happily let an assistant compare prices for them. Very few will let it place the order. So we work where the behaviour is — discovery, comparison, and the product data both of those run on — and we build transactability as readiness rather than as the pitch.
Challenge
one
“Our competitor gets recommended and we don’t, and nobody can tell us why.” We start with the question that can actually be answered: did the retrieval system fetch your page at all, and could it parse what it found? Your own server logs settle that. Retrieval audit Crawler posture Source corpus
Challenge
two
“Organic traffic is down and the dashboards say nothing changed.” Because nothing changed on your side. The behaviour changed. We rebaseline what good looks like, move measurement onto presence, and rebuild the demand mix around the new shape. Zero-click Measurement rebaseline Demand mix
Challenge
three
“Our product pages are invisible to whatever is doing the recommending.” Product pages are consistently the least machine-readable page type. Where a product comes from a properly built feed, the brand, image and price come through intact every time. Feed engineering Machine readability Conversational attributes
Challenge
four
“We can see AI referral traffic. It’s small. Should we care?” Yes, but not for this quarter's number. It converts better than almost anything else you have and it's compounding, while the channel it replaces contracts. Channel strategy Investment case Halo measurement
Challenge
five
“Quick commerce is a real channel now and we manage it like a spreadsheet.” Listing and category-page optimisation, sponsored search, first-party targeting, and attribution that runs from impression to doorstep on the same day. Retail media Quick commerce Marketplace
Challenge
one
“Our competitor gets recommended and we don’t, and nobody can tell us why.” We start with the question that can actually be answered: did the retrieval system fetch your page at all, and could it parse what it found?
Challenge
two
“Organic traffic is down and the dashboards say nothing changed.” Because nothing changed on your side. The behaviour changed. We rebaseline what good looks like, move measurement onto presence, and rebuild the demand mix.
Challenge
three
“Our product pages are invisible to whatever is doing the recommending.” Product pages are consistently the least machine-readable page type. Where a product comes from a properly built feed, brand, image and price come through intact every time.
Challenge
four
“We can see AI referral traffic. It’s small. Should we care?” Yes, but not for this quarter's number. It converts better than almost anything else you have and it's compounding. Payback is months, not weeks.
Challenge
five
“Quick commerce is a real channel now and we manage it like a spreadsheet.” Listing and category-page optimisation, sponsored search, first-party targeting, and attribution that runs from impression to doorstep on the same day.
Structured interventions
What we build.
The way in to this capability is the Retrieval Audit. Everything below is what happens after it — defined, bounded commercial engagements.
AI Retrieval & Machine-Readability Audit
The way in, and the highest-integrity thing we sell, because every input is deterministic and belongs to you. Did the retrieval systems fetch your pages? Could they render and parse them? What’s the ratio of crawling to actual referrals? Is your firewall quietly blocking the very systems you want to be read by? Page-type readability scored by template, homepage tested as a definition of the organisation.
Inquire about this scopeAI Visibility Baseline & Tracking
A library of the questions your buyers actually ask, built by stage — discovery, comparison, evaluation, validation, objection — and asked many times over across the major assistants. Reported as how often you appear, and how that compares to your competitive set, shown as a trend with its variation. Never as a rank. Never as a single score out of a hundred. Paired with the signals that move first: branded search and direct traffic.
Inquire about this scopeSource Corpus Programme
Getting into the sources each system actually draws on — refreshed quarterly, because the mix genuinely moves. Plus original first-party data and real first-hand expertise, which is the one thing a model can’t already produce for itself.
Inquire about this scopeProduct Data & Feed Engineering
Where machine-mediated commerce actually pays today. Feed integrity — identifiers, live pricing and availability matched to real inventory, clean imagery, correct categorisation — plus the newer conversational attributes most brands haven’t touched: question-and-answer pairs, related-product relationships, links to manuals and spec sheets, and variant hygiene.
Inquire about this scopeAgent Interface Layer
Making the brand transactable and correctly presented when a machine arrives. Capability declarations, a functional description of what you sell written for a system rather than a shopper, the icons and naming that control how you appear inside an assistant’s product card, server-side rendering, and firewall rules that challenge rather than silently block.
Inquire about this scopeRetail Media & Quick Commerce
Listing and category-page optimisation, sponsored search, first-party targeting and same-day attribution across the platforms where purchase decisions are increasingly made.
Inquire about this scopeEarned, Creator & Community
The mechanism everything above depends on, offered as a service rather than assumed as an outcome. Digital PR into publications that are genuinely cited, creator programmes run with disclosure built in from the brief, and owned community presence on the platforms that show up disproportionately in AI answers.
Inquire about this scopeSearch & Paid Demand
The unglamorous majority of the volume. Technical fundamentals, which matter more now rather than less, plus paid search, shopping and performance.
Inquire about this scope
how we measure
Our method,
published.
A large share of people working in this field say they don’t trust the data available to them. In that environment, publishing how you measure is worth more than any claim about results. So here’s ours.
What we treat as fact.
DETERMINISTIC & OWNED
Things that are deterministic and that you own. Whether the retrieval systems actually fetched your pages, and what they got.
How much crawling you’re absorbing relative to referrals you receive. Referral segmentation in your own analytics. And branded search and direct traffic, which usually move before anything else does.
What we treat as statistical.
TRENDS & VARIATION
How often you appear across a defined library of questions, asked many times over. How that compares to your competitive set. How you’re described — accuracy, framing, consistency.
And how the conversation develops, because brands that are absent from a first answer often appear by the third.
All of it reported as a trend, with its variation shown.
What we don’t report at all.
INTEGRITY & LIMITS
A rank position in an AI answer. Search volume for prompts, presented as measured fact. A single visibility score out of a hundred.
And a claim that one specific change caused one specific movement, when the systems involved don’t behave identically twice.
Before we quote you.
CATEGORY-SPECIFIC BASELINE
How often any brand gets cited varies enormously by category. Travel and automotive have real surface area. Professional services has very little.
If you’re in a category where citations are scarce, a citation-count target is the wrong measure and we’ll propose a different one rather than sell you a number we know will disappoint.
our standard
If a metric wouldn’t stand up to technical verification or statistical scrutiny, it doesn’t belong in our reporting.
We build systems designed to measure real movement, not comfortable illusions.
frequently asked
Questions
& answers.
What e-commerce and growth leaders usually ask before commissioning a retrieval audit.
Retrieval and readability fixes can move within weeks, because they remove a blocker. Presence in answers compounds over months. Anyone promising faster is measuring something else.
All the major ones, as standard. Platform share has moved dramatically in a short time — the leader has lost a large chunk of it and challengers have multiplied theirs. Optimising for one assistant is a defect, not a strategy.
A lot of it is, and we’ll say so. The AI features everyone is chasing run on the same underlying systems that rank pages. What’s genuinely new is the retrieval layer, the entity layer and the feed layer. The rest is technical fundamentals that matter more than they used to.
We do it. Feed engineering is where machine-mediated commerce actually pays today, and it’s mostly unglamorous data work that nobody at scale wants to own.
next step
Find out
why you’re invisible.
Three to four weeks. Fixed fee. Every finding backed by evidence you own and can check without us.