unvibe

The audit and the score

An Unvibe checkup reads a live website the way AI crawlers do and answers one question: why could AI assistants and agents fail to understand, trust, recommend or use this business? It gives a short answer, a ranked list of findings with the evidence behind each one, the customer questions AI can and can't answer from the site, and a score out of 100.

What a checkup reads

  • The homepage's raw HTML, without running JavaScript, the way GPTBot, ClaudeBot and PerplexityBot read it.
  • The homepage as each AI crawler: GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Googlebot, with their real user agents. This catches CDNs, firewalls and bot protection that turn AI crawlers away.
  • The pages customers ask about: pricing (or the menu), contact, about, help or FAQ, and where the main sign-up, buy or booking link goes. Unvibe follows the homepage's own links, and tries the usual address when there is no link.
  • robots.txt, rule by rule for each AI crawler, and the sitemap.
  • For shops, one product page, found from the homepage links or the sitemap.
  • An AI's reading of exactly those pages, starting with the niche (see below), then what the business offers, its reasons to choose it, anything that contradicts itself, and which customer questions the pages answer. Every quote the AI gives is checked against the pages; quotes that aren't there are dropped.

How the checkup works out your niche

Before it judges anything, the checkup works out which market the business is in, from its own pages: the headline, what is offered and to whom, the prices, the names of the pages in the sitemap, and the structured data. The report's How AI reads card shows:

  • The market, category, audience and area (for businesses that serve a place), with the sentence it was read from.
  • How sure the AI is. "Sure" needs a sentence on the site that states the market. When the AI can only guess, the card says what the pages leave out, the category check fails, and the audit asks you to say it plainly instead of filling the fixes with a guess.
  • What buyers in that market compare, who they weigh you against, and how they ask AI. These come from what the AI knows about the market, not from your pages, and the card says so. You can track any of the buyer requests in AI mentions.

If you left the niche empty, or it still says "Business", the first checkup the AI is reasonably sure about fills in the niche and audience for you, and your questions, mentions and fixes use them. A niche you typed yourself is never changed; if the AI reads your pages differently, that is a finding.

The audit

The audit is a short list of findings, most important first. Each one says:

PartWhat it gives you
What we sawThe problem on this site, with quotes and values from the live pages
Why it mattersHow it makes AI fail: it can't read, understand, trust, recommend or use the business
Backed up byWhere your AI mentions, fact check, competitors or tracked prompts show the same problem
What to changeThe exact change, often with a ready snippet built from the site's own facts
How Unvibe will check itWhat the next checkup looks for to confirm the change worked

Every finding is critical (likely to make AI fail) or an opportunity (would make AI more confident), and is ranked by its likely impact and the effort to fix it. What already works is listed separately, so it stays that way.

The audit reports one cause once. A page that is empty without JavaScript is one finding, not a dozen symptoms, and the checks that depend on reading the page wait until it can be read. An FAQ that is missing and customer questions that go unanswered are one finding.

What the audit looks for

  • Can AI reach it? robots.txt or a noindex tag blocking AI crawlers; a firewall or CDN turning them away; text that only appears after JavaScript runs; a canonical tag pointing to another page or site; a sitemap that is missing or leaves out key pages.
  • Can AI understand it? A title, description and headline that don't say what the business is or who it's for (including template titles left by site builders); structured data that is missing, broken or describes the page instead of the business; the business name not used on the site.
  • Can AI trust it? Prices or claims that contradict each other between pages or between the page and its structured data; no proof AI can cite (customer counts, named testimonials, review scores, certifications); nobody named behind the business; contact details missing or only in hidden structured data.
  • Will AI recommend it? No readable price; no concrete reason to choose it (judged against what buyers in your market compare); no page comparing it with the alternatives buyers weigh it against; questions buyers in your market ask left unanswered.
  • Can an agent use it? A main button an agent can't press (not a real link), a vague label, a sign-up or booking link that leads to an error or a login wall, no FAQ or help to find answers in.
  • Shops: product pages without product data and a price; no shipping and returns pages linked.
  • Local businesses: no address or opening hours as text, or missing from the LocalBusiness data.

Customer questions

The audit lists the questions customers ask AI about a business and says for each whether the pages answer it, with the sentence that does, or which finding fixes it. The basics are the same for everyone: what it does, who it's for, what it costs, how to start, why trust it and how to get in touch, plus delivery and returns for shops and where and when for local businesses. The rest are four to six questions that buyers in your market ask before choosing: a dentist is asked whether it accepts a patient’s insurance, invoicing software whether it handles VAT. The next checkup asks the same questions again, so an answer you add can be confirmed.

The score

The score out of 100 comes from 17 checks across three areas. It is calculated the same way as always, so it stays comparable over time. Some findings, such as contradictory prices or a wrong canonical tag, aren't part of the 17 checks: fixing them shows as verified work rather than a higher score.

The three areas

AreaShare of the scoreWhat it measures
Understood by AI40%Whether an AI can tell what the business is, who it is for, what it costs and why to pick it
Readable by crawlers35%Whether AI crawlers can reach and read the page, and find the rest of the site
Usable by agents25%Whether an AI agent can find prices, the main action (sign up, buy or book) and help on its own

Each check has a weight. An area's score is the share of its weight that passes, and the overall score combines the three areas at 40%, 35% and 25%.

The 17 checks for every business

Readable by crawlers

CheckWhat passes
Page descriptionA meta description that says who the business is for
Main headingOne clear main heading that says what the business does
SitemapA sitemap that crawlers can find
robots.txtrobots.txt lets AI crawlers such as GPTBot and ClaudeBot in
AI crawlers get the pageThe CDN, firewall or bot protection serves AI crawlers the same page as browsers
Structured dataJSON-LD that describes the business: what it is, what it offers and its prices
Link previewsA title, description and image for link previews (Open Graph)
Real buttons and linksClickable things are real buttons and links, not plain boxes

Understood by AI

CheckWhat passes
What it isThe first line of the homepage says what the business is
Who it's forThe audience is named near the top of the page
CategoryThe page says plainly what kind of product it is
Reasons to pick itConcrete reasons to choose it over alternatives
Price on the pagePricing is written as text on the page, not only in a pop-up
ComparisonA page that compares the business with its closest alternatives

Usable by agents

CheckWhat passes
Prices linkA link to prices (pricing, menu or products) in the main menu
Main buttonA clearly labelled main action an agent can find: sign up, buy, book or contact
HelpA short FAQ or help section

Extra checks by kind of business

The checkup works out what kind of business the site is and adds the checks that fit.

KindCheckWhat passes
ShopsProduct dataA product page has Product structured data with a price
ShopsShipping and returnsThe site links to its shipping and returns pages
Local businessesBusiness detailsLocalBusiness structured data, plus a phone number and an address (and ideally opening hours)

Why a page looks empty to AI

When the crawler gets little or no text, Unvibe names the most likely cause:

  • JavaScript rendering: the text is added by JavaScript after the page loads, and AI crawlers don't run it. Fix: server-side rendering, static generation or pre-rendering.
  • AI crawlers blocked: browsers get the page but AI crawlers are blocked or get a different page, usually by a CDN, firewall or bot-protection setting.
  • Blocked by robots.txt or noindex.
  • Thin page: readable, but too little text to explain the product.
  • Page error: the homepage returned an error.

Fixing and verifying

Copy a finding for your coding agent (or let the agent read it over MCP), deploy, then mark it as fixed. The next checkup tests the live site: a finding counts as fixed only once the audit no longer raises it, and a check only once it passes. Fixes that are confirmed are kept as work completed and appear in the weekly report and on shared reports. If the live site still shows the problem, the checkup says so.

Questions

How often should I run a checkup?

After each deploy that changes your homepage, robots.txt, sitemap or CDN settings. A checkup that fails, for example because the site is down, doesn't count toward your allowance.

Does the checkup look at every page?

The checkup reads the homepage and the pages customers ask about (pricing, contact, about, help and the sign-up target). The page audit reads up to 15 pages from the sitemap and flags the problems that stop AI using each one, such as an empty page, a duplicate title, a canonical tag pointing elsewhere, or a pricing page with no price.

What happens when my score reaches 100?

AI can read all of it. The next step is being recommended: see AI visibility and rank tracking.

Last updated September 29, 2026 · View as Markdown · All docs as one text file

See what ChatGPT sees on your app.

Free to start. Fixes ready for your coding agent.

Check my app