How Fecto works it out

Fecto charges only for a result it can prove, so it has to be plain about three things: which questions it tracks, how it decides what work to do, and how it measures the result. This page says how each one works today.

1. How the questions are chosen

AI assistants answer questions, so Fecto keeps a short list of the questions your buyers actually ask.

  • You confirm every question. Your agent proposes questions from your site, your Search Console queries or what you tell it. None counts until you confirm it on your dashboard. Questions are written as generic buyer questions and kept private to your account. Questions with an email address, a phone number or a link are refused.
  • Each question has a topic and says who is asking:
    • exploring the problem
    • ready to choose
    • comparing options
    • naming a competitor
    • naming your business
  • Questions that name a business never count in a rate. If a question names you or a competitor, Fecto still tracks it, but leaves it out of every mention rate and every measure you could be charged on. It applies this rule to the words of the question, whatever label it was given. Those questions show how assistants treat you once someone already knows your name. They don’t show whether assistants find you unprompted.
  • Same name, different product. You can list other products that share your name. An answer that names one of them isn’t counted as naming you. When an answer names both and Fecto can’t tell which one is meant, it doesn’t count either, so the rule errs low.
  • Mention rates use fixed questions. When Fecto measures how often AI answers name a business, it asks a fixed set of questions written from the business’s category and area (“best bakery in Austin”), never from one customer’s private questions. Your own questions run privately for your site. A separate control set, with well-known brands in unrelated categories, shows how much the assistants drift on their own. Fecto asks ChatGPT, Perplexity and Claude. It never samples Gemini.
  • Two ways to change a question.
    • Rewording keeps the question and its history, and its version number goes up.
    • Replacing it with a different question starts its history again from zero.
    A plan you’ve already accepted keeps the questions, and the wording, it was accepted with.

2. How the work list is made

From the first scan, before any card or plan, Fecto keeps one list of the work for your site. Each item says:

  • what to do
  • why, with the evidence
  • which result it serves: visits from search, visits from AI assistants, or enquiries and bookings
  • which page
  • what kind of work it is: a new page, a page edit, a technical fix, outreach or community

Three kinds of research feed it:

  • The scan. Fecto reads your live site the way people, search engines and AI assistants do. Each check that fails, and that a known fix serves a result, becomes an item. The public report shows these items to anyone.
  • The pages AI cites. Your confirmed questions are asked in AI assistants, by Fecto’s own panels or by your agent in the assistant you already use, and the answers come back with their citations. Fecto reads every cited page itself. A page that recommends others but not you becomes an outreach item: ask to be included, never pay for it. A forum thread or video becomes a community item: an honest answer from your own account, saying who you are.
  • Your pages beside the cited ones. For each question, your agent reads the pages AI cited next to your own page for it, on your computer. It reports two things only:
    • a claim those pages make that your page lacks
    • a fact your page gets wrong against your confirmed business record
    Every finding must name the pages it rests on, and the few words your page should state once fixed. Fecto runs no AI model of its own for this.

Items are ranked by a fixed rule, so you can check the order yourself:

  • A scan item scores 60 for a serious problem, 35 for one to improve and 20 for one shown for information, plus 5 for each result it serves.
  • A page comparison scores 55 for a wrong fact and 40 for a missing claim, plus 5 for each AI engine (up to 3) that cited the pages compared.
  • An outreach or community item scores 20, plus 8 for each engine (up to 3) that cited the page, plus 5 if Fecto can check it.

Each item has a status:

  • Suggested: on the list, not started.
  • In progress: your agent is working on it as a step of an accepted plan.
  • Handed in: your agent reported it done, and Fecto graded the step.
  • Passed Fecto’s check: Fecto confirmed it on your live site, by a fresh scan, by fetching the page and finding the agreed words, or by finding your name or link on the third-party page.
  • Didn’t pass: the live check still found the problem.
  • No longer suggested: a newer round of research stopped suggesting it.

Your agent still gets the work one step at a time, and each step has to pass Fecto’s checks before it closes.

3. How the result is measured

  • Rules are frozen when you accept. Freezing locks the check ids and rule versions, the starting point, the target and price bands, the deadline, and, for AI goals, the question set. Fecto can change its rules later. Your plan keeps the ones it was accepted with.
  • Growth results are counted by Fecto, not by your agent. Visits from search, visits from AI assistants, and enquiries and bookings are counted by the Fecto tag on your live site, or by signed events from your backend if it sends them. The count covers the 30 days before the deadline, once per visitor per day, and the result is judged once, at the deadline.
  • The starting point. Once the tag has 28 days of data, the start is its own 30-day count. Until then, it is the higher of the tag’s count scaled to 30 days and a number you chose to share. A new or quiet site counts its growth from at least 30 search visits, 5 AI visits or 3 enquiries.
  • Growth is relative, not points. +20% means 20% more than the starting count: from 50 to 60, not 20 percentage points. The standard price bands:
    • +20%: $30
    • +50%: $60
    • +100%: $100
    A band is priced only if chance alone would not reach it: its count must be above the 90th percentile of normal ups and downs for that starting point. Your agent can agree a lower cap (from $10), steps of 10%, or a deadline between 30 and 120 days. Every price stays inside those limits.
  • Every collection is counted, including the failed ones. Your plan page shows how many collections were planned, how many have happened so far, how many gave usable data and how many failed. For a growth result, a collection is a daily count from the tag. For a mention rate, it is an AI answer.
  • Mention rates. A mention rate is the share of answered questions whose answer names you. Failed asks (an engine errored or timed out) are kept out of the denominator, shown separately, and never re-asked to get a better sample. Rates come with a 95% range. A goal paid on mentions measures the rise in percentage points, and the rise must beat the control set’s drift with 90% confidence. Fecto never sells a single AI answer as a result.
  • Delivery before results. Before it reads a result, Fecto checks that the changes your agent handed in are live on your site, and says which aren’t. A reading is never presented as the effect of a change that isn’t deployed.
  • Everything on one timeline. Your plan page lists, in date order:
    • each change and when it went live
    • every check
    • anything else that changed in the same period, such as a price change, an ad campaign, a redesign or an outage
    The other changes are listed because they can move a result too. They explain a result. They never decide it.
  • No AI decides the verdict. The verifier reads only what was frozen, measures with Fecto’s own tools, and decides by fixed rules. The agent that did the work never grades it.

When you pay $0

  • The count at the deadline is below the first band.
  • A part of a per-unit goal doesn’t pass, or Fecto couldn’t check it. You pay only for the parts that pass.
  • You cancel a plan. Cancelling costs nothing.
  • Saving a card. It costs $0 and starts no plan or deadline.

Whatever a result reaches, one site is never charged more than $100 a month. If you think a result is wrong, you can dispute it for 30 days, and collection pauses while a person at Fecto reviews the checks.

Start

Check your website for free, or read about Fecto.

Questions: support@fecto.ai