The terms, defined by the numbers we use
What our scan measures, what our calculators assume and how our prices are built, one term at a time. Where a definition quotes a weight or a figure, it is the one the product actually runs on.
Measuring AI answers
- AI visibility score
A score out of 100 for how often AI assistants name a business when a customer asks them for one. In the AIGrow scan, being named is worth 45 points, being described 15, reach across questions 20, and the machine-readable basics on the website 20. Questions that fail to get an answer are left out rather than scored as zero.
Run the free scan- Buyer question
A question a customer would type into an AI assistant while choosing a supplier, such as who the best dentist in a named town is. The AIGrow scan derives five of them from the business’s own website, in the words a customer would use, and puts each one to up to four assistants.
- Probe
One buyer question put to one AI assistant, with the businesses it names recorded. The assistant is asked the customer’s question rather than whether it knows the business, because a model asked about a named business almost always says something agreeable. The answer is then read for the names it contains.
- Failed probe
A probe that returned no usable answer, because the provider timed out or refused. The AIGrow score leaves failed probes out of the arithmetic instead of counting them as absences, since a provider outage is a fault in the measurement and says nothing about whether the assistant knows the business.
- Named
The business’s name appears in an assistant’s answer to a buyer question. It is the axis that decides the AIGrow score: the share of answered probes that name the business is worth 45 of the 100 points, because everything else refines a business the assistant already knows exists.
- Described
The assistant said something about the business beyond its name, such as what it does or why it suits the customer. In the AIGrow score it is the share of mentions that come with a description, worth 15 points, because a bare name is a weaker recommendation than one the assistant can explain.
- Reach
How many of the different buyer questions produce a mention at least once, across all the assistants asked. It is worth 20 points in the AIGrow score. A business named every time for one question and never for the other four has a narrow reach, however healthy its mention count looks.
- Machine-readable basics
The facts an assistant can read from a website without guessing. The AIGrow scan awards up to 20 points for them: Organization or LocalBusiness schema 8, or any other schema 3, a phone number in the header 4, a street address 3, opening hours 2, and a description of at least 50 characters 3.
- Degraded scan
A scan that could not measure the business, so it reports no score. It happens when no assistant answered, or when buyer questions could not be derived from the site and a generic set was used, which cannot produce a mention. A degraded scan is never published, monitored or reported as a zero.
- Frozen question set
The same buyer questions, asked again on every run of a monitor. Questions that changed every month would measure the questions rather than the business, so a monitor keeps the set it started with and skips deriving new ones. A change in the score then means the answers changed.
- Mention and citation
A mention is the business’s name in the text of an answer; a citation is a link to a source shown beside it. The AIGrow scan measures mentions, because citation coverage is inconsistent across assistants and what matters to a buyer is whether the name appears in the answer at all.
- Run-to-run variation
How much an assistant’s answer changes when the same question is asked twice. In our calibration on twelve local buyer questions, only 16% to 26% of the businesses named across two runs of the same model appeared in both. One answer is a sample, which is why a scan asks several questions of several assistants.
- Rivals
The other businesses the assistants name in answer to the same buyer questions. The AIGrow scan counts how often each one appears across the answers and lists up to six, most frequent first. The scanned business never appears in its own list, even when an assistant shortens its name to one word.
Costs and automation
- Loaded cost
What an employee costs the business in total: salary plus employer payroll taxes, benefits, paid leave and equipment. It is the figure that compares fairly with a monthly subscription. For a full-time receptionist in the US, the typical loaded cost we use as an anchor is $3,170 a month.
Work out a role’s cost- Anchor figure
A typical figure printed where the reader’s own is unknown, and always labelled as one. Our role anchors are loaded monthly costs for a US small business: receptionist $3,170, tier-one support agent $3,750, SDR $5,400, and outsourced bookkeeping $2,000. Our calculators replace the anchor as soon as a real figure is entered.
- Close rate
The share of enquiries that become a sale. In the missed-call calculator it turns unanswered calls into lost revenue, and it is the input that moves the result most. We assume 25% when none is entered; a rate taken from your own booking or sales records makes the result yours.
Cost your missed calls- A month, in the arithmetic
Our calculators count a month as 4.33 weeks when the input is hours a week, and as 22 working days when the input is a daily count such as missed calls. A year is twelve of those months. Both conventions are printed beside the result that uses them, so it can be checked by hand.
- Cost per answered call
What it costs to have one inbound call picked up and handled. A live answering service typically charges $8 to $12 a call on per-minute pricing, while an AI-handled call typically costs about $0.40. It matters because 62% of calls to small businesses go unanswered, according to a 2026 small business voice study.
See the AI receptionist- Cost per resolution
A fee charged for each customer conversation an AI resolves without a person. Published 2026 pricing puts it at $0.99 on Intercom, $1.50 on Zendesk and $2 on Salesforce. The AIGrow support agent charges no per-resolution or per-seat fee: it runs at $79 a month after a build from $1,490.
See the support agent- Speed to lead
The time between an inbound enquiry and the first reply from the business. An inbound lead is worth chasing for about five minutes, and after that it cools quickly. A web form collects enquiries around the clock while someone watches it only in working hours, and that gap is what an SDR system closes.
See the SDR system- Month-end close
The work of reconciling a month’s transactions so the books can be signed off. By hand it typically takes three to five days; when transactions are matched continuously it shrinks to hours. The accountant still signs off the accounts, and automation takes the typing and matching that come before the signature.
See the reconciliation system- Build price and running cost
The two numbers every AIGrow system is priced on: a build, paid once, and a monthly running cost for hosting, model usage and monitoring. Both are fixed in writing before work starts, with no per-seat fee and no minimum term. The role calculator counts the build in the first year only.
- AI agent
Software that watches for a trigger and acts without being asked: it reads the incoming event, decides against your rules, does the work in your own tools and logs every step. An assistant waits to be asked. When an agent meets a case its rules do not cover, it escalates instead of guessing.
How an agent works- Escalation path
The route by which an automated system hands a case to a person, with the context attached. Every AIGrow role system has one, and each role page states what stays human: an upset caller for the receptionist, an angry customer for support, and the sales call itself for the SDR.
- Process audit
A fixed-price review that measures what a team does by hand and what it costs, before anything is built. The AIGrow audit costs $290 and returns a written document in three to five days. If a build follows, the full $290 comes off its first invoice, so the audit is free for anyone who goes ahead.
See the audit
Models and tokens
- Token
The unit a language model reads, writes and bills by: a word, part of a word or a punctuation mark. Providers price usage per million tokens, with input and output charged separately. Because each model family splits text its own way, the same page costs a different number of tokens on different models.
- Tokenizer
The rule a model uses to split text into tokens. It differs between model families, so token counts and per-token prices are not comparable across them. In our own test, one English paragraph counted 577 tokens on Claude Sonnet 5 and 381 on an OpenAI tokenizer: the cheaper price per token can still be the dearer page.
See the numbers on your own business
The free scan puts your customers' questions to AI assistants and scores the answers with the weights defined above.