Do you only check online shops?

No. We check any website. Before we grade anything we read your site and work out what you actually are — whether you sell products, services, software, content or advice, whether you serve one town or the whole world, and how someone becomes a customer. You are then graded against that, not against a shop template. A law firm is never asked for barcodes.

Why don't AI assistants recommend my business?

Usually one of five things: your site tells the assistants' crawlers to stay away, your pages carry no machine-readable data about what you offer, there is no identifier or location an assistant can confirm you by, the details it filters on are missing, or your titles and headings are too vague to match somebody's question.

What does Chorrus fix?

We publish what you offer in the shape a machine reads: identifiers and attributes for products, service and location details for a local business, plans and comparisons for software. We rewrite titles and descriptions into the words your buyers use. We never invent a fact about your business, and every change can be undone.

What is the 360 view?

We read the whole business before measuring it — what you offer, where you serve, how someone converts, who your buyers are and what decides their choice — then follow the result past the recommendation: were you named, did that produce visits, did those visits engage, and did they convert. That tells you which of the four things to do: fix upstream, improve how you are described, pivot the page, or double down.

Why does it need watching after it's fixed?

Every assistant keeps changing what it reads and how it weighs it. A business that fixed things once and stopped looking can slide back out of the answers without noticing. We re-check on a schedule and show where you are named, who is named instead, and what has opened up.

How it works

What's stopping you, what we fix, what we keep watching.

When someone asks an assistant for what you do — a product to buy, a firm to hire, a tool to try, a shop to visit — it doesn't walk around your business the way a person would. It reads your site and names the ones it can confidently understand. If yours can't be read, it can't be named, and you never see the sale, because there was never a visit.

Any business, not just shops

This used to be a tool for online shops, and everything that wasn't one came out at zero with a list of demands it could never satisfy — submit a product feed, add barcodes, mark up your catalogue. A law firm has no catalogue. That wasn't a harsh score, it was the wrong question, and it's been fixed.

There is now one pipeline and no dropdown for "business type", because the type was never the useful thing. What varies is a handful of independent facts, and we resolve them from your own site before grading anything:

What you offer — products, services, software, content, advice, or any mix of them. Mixes are the normal case, not an edge case: a hobby shop that also runs classes is graded as both.
Where you serve — one neighbourhood, a region, a country, or everywhere. This decides whether "near me" questions are even asked on your behalf.
What a buyer gets pointed to — a product, a service line, a branch, a named person, an article, a plan.
How someone becomes a customer — checkout, quote, booking, trial, demo, phone call, walking in.
What decides the choice — specification and price, or reputation, credentials and proof. Nobody picks a surveyor the way they pick a cable, and the rubric shouldn't pretend otherwise.

Two absences are kept strictly apart, and it matters more than it sounds. Not applicable is an answer: a software company has no premises, and that is a fact about it, not a hole in it. We couldn't tell is a flag: we keep our best reading, we mark it as weak, and we don't let it drive a finding. An early version of this read a testimonial byline as an address and told a company with no offices to mark up its premises. Now a weak reading has to earn its way past a confidence threshold before anything is built on it.

Why it matters

People are moving from searching and scrolling to just asking — on ChatGPT, Gemini and the rest, all at once. Two things follow from that. You're either in the answer or you're nowhere: the assistant names a few businesses, often one, and there's no second page for anyone to find you on. And it builds on itself: the ones that can be read get named again and again, while the ones that can't stay unnamed. The big platforms, marketplaces and directories were wired in from the start. This is the rare channel where knowing your niche beats outspending someone — but only if a machine can read what you do.

The size of it, in plain numbers: in a single year, more than twelve times as many shoppers reached U.S. retail sites from an AI assistant as the year before. Half of U.S. shoppers used AI to help them buy something last year, and about two in three say they will this year. Roughly one person in ten on earth opens ChatGPT in a given week. Retail is where this got measured first, not where it stops — the same habit now picks contractors, agencies and software. None of it is a forecast; it already happened.

Sources: Adobe Analytics (2025) · PartnerCentric consumer survey (Dec 2025, n=1,004) · OpenAI (2025).

What breaks

An assistant reads your site the way a machine does, not the way a customer does — and when it can't, nothing on your site looks broken. Here's what actually stops it, in the order it usually bites:

Your site turns the assistants away. There's a small file most businesses never look at that tells automated visitors where they may go. If it says no to the AI ones, they can't read a single page — and nothing about your site looks wrong to you.
What you offer exists only as prose. Sites publish a hidden summary for machines to read. Without it, an assistant is guessing from paragraphs, and it would rather recommend somebody it doesn't have to guess about. For a shop that's the product block; for a service business it's the services themselves; for software it's your plans.
There's nothing it can confirm you by. For a product, that's a barcode or part number. For a local business, an address and hours a machine can read. For a firm, the licences, registrations and named people that prove you are who you say. No confirmation, no confidence — so it names somebody it can verify.
The details are missing. Size and material; areas covered and hours; what the plan includes at which tier. Those are what an assistant narrows a list down by, and anything without them gets narrowed out.
The wording is too vague to match. "Bundle of 3" tells a machine nothing, and neither does "Solutions". Assistants match on titles and headings first, so a vague one is skipped before anything else is considered.
The proof your category runs on isn't there. Where buyers choose on trust rather than spec, missing reviews, credentials, case studies and named expertise are the whole problem — and they're invisible on a scorecard built for catalogues.

The 360 view

Before we grade anything, we answer the same fixed set of questions about every business we read — the same ones, in the same shape, whether you're a dive shop or a design agency. Eight groups: what the business is and how it makes money; what it offers and how that's organised; where it serves; how someone converts; who the buyers are, what brings them and what they're anxious about; what proof and authority you carry; what an assistant can actually retrieve from your site; and who else is already in your category's answers.

It's deliberately fixed rather than adaptive. An interview that asks a software company different questions than a shop produces profiles that can't be compared to each other — and comparison is the entire point of everything downstream. Depth comes from asking all of it every time.

Those answers are what generate the questions we then put to the assistants: real buyer questions, in buyers' words rather than your marketing vocabulary, and localised only when we actually know where you serve. And they're kept apart from what your site literally says — we can always show you the sentence we read and the conclusion we drew from it, and never confuse the two.

Past the recommendation

Being named is the first stage, not the finish. Connect your analytics and we follow the same question all the way through: were you named in the answer, did that produce clicks, did those visitors engage or bounce, and — where you have it wired up — did they convert. Where the chain breaks tells you what to do, and there are only four answers:

Not named → fix upstream. The work is in what an assistant can read about you.
Named, few clicks → improve how you're described in the answer. You're there; you're just not compelling in it.
Named, clicked, then lost → pivot the page. The recommendation is landing and the page isn't closing.
All four healthy → double down, and go after the questions next door.

You get one recommendation at a time, not a list of five. Five at once means that when something moves, nobody can say which change did it — and knowing which change did it is the thing that makes the next recommendation better. If we can't show the working behind a number, we don't show the number: a stage we couldn't measure says "not measured", never zero.

What we fix

We publish what you offer in the shape a machine reads — identifiers and attributes where you sell products, service and location detail where you serve an area, plans and comparisons where you sell software — and rewrite titles and descriptions into the words your buyers actually use. Two rules never bend: we never invent a fact about your business — a missing barcode, licence number or measurement comes back to you to supply, never guessed — and every change can be undone, one item or the whole batch.

What we watch — and what it teaches

This is the part most businesses skip, and it's the part that decides whether the work holds. Every assistant keeps changing what it reads and how much weight it gives each thing. A business that fixed everything in the spring and stopped looking can be back out of the answers by autumn without a single sign on its own site — no drop in enquiries it can explain, no error, nothing to notice.

So we keep asking. On a schedule, we put real buyer questions to each assistant and record where you're named, who's named instead, and which questions have opened up since last time. When something slips, you hear it from us rather than from a quarter of quiet.

And because we're sitting in the measurement seat while you make changes, every fix gets timestamped against the numbers it was meant to move. Over time that answers the question you actually care about — if I do this, will it work? — with evidence rather than opinion. We hold that to three tests before calling anything a win: enough runs before and after to mean something, a check that your untouched pages didn't drift the same way anyway, and a comparison against similar businesses that changed nothing over the same weeks. Results come back as worked, no measurable effect, or not enough evidence. A "didn't work" is a finding we'll tell you, because it saves you the next month.

Category comparisons and trend calls stay switched off until enough businesses in a category are being measured to make them honest, and nothing is ever published that could expose one competitor's private numbers. When something is being withheld for that reason, the report says so rather than showing you noise. And we never quote you how many people ask an assistant a given question — nobody publishes that. What we can tell you is what's rising, roughly how much warning you have, and whether you're named yet.

Check your site — free