Methodology

This page documents how NoCall turns thousands of user reports into a reliable assessment for every phone number. We explain the exact risk-score formula, the danger levels, how we classify categories, how our artificial-intelligence analysis works and where the data comes from. We publish this methodology so that anyone can understand —and question— the decisions the system makes.

How the risk score is calculated

Every number in our database is given a risk score between 0 and 100. It is not a subjective judgement: it derives from a deterministic formula that is recomputed every time a new approved report arrives.

risk = min(100, reports × 5 + (verified ? 30 : 0) + min(10, external opinions × 2))

Each approved report adds 5 points. If our team has verified the number as confirmed spam, 30 points are added. Public opinions collected from other websites add 2 points each, 10 at most. The result never exceeds 100.

The verification factor exists to distinguish a number with many recent reports (which could be a one-off false positive) from a number a human reviewer has actively confirmed as abusive. That is why 20 unverified reports (100 points from reports, capped) and 14 verified reports (70 + 30) can reach similar scores by different paths.

When a number has been analysed by our AI and has approved community signals, the detected danger level also acts as a floor for the score: a number flagged as critical will never show below 90, a high one below 70, medium below 45 and low below 20, even with few reports. This way a clearly fraudulent pattern is not undervalued just because it is recent.

Danger levels

From the numeric score we classify each number into one of four danger levels. These are the same identifiers our database uses internally:

Spam categories

Every report and every number is classified into one of seven categories. The category determines how the number is presented in the directory and is derived both from the user's report and from the later automated analysis:

SPAM — Generic, unwanted commercial calls that do not fit a more specific category.
TELEMARKETING — Telephone sales campaigns, usually telecoms or energy, that persist despite rejection.
SCAM — Scams and fraud, including identity impersonation (banks, public bodies, fake tech support). The most serious category.
DEBT — Debt-collection and recovery calls, often aggressive or aimed at the wrong person.
HARASSMENT — Repeated calls intended to harass, intimidate or deliberately disturb.
SURVEY — Unsolicited phone surveys, polls and market research.
OTHER — Any other type of unwanted call that does not fit the above.

Artificial-intelligence analysis

Besides the numeric score, an AI worker analyses numbers using the published comments and reports. That includes reports and comments from NoCall users that an admin has approved, but also published comments that do not come from NoCall users and that nobody has reviewed. Content still pending review is not used.

For each analysed number, the AI generates a structured set of information shown on the number's page:

The fields the analysis produces are as follows:

Description — A natural-language summary of who seems to be behind the number and what they want.
Complaint patterns — The most repeated complaint reasons across reports (for example, calls at unsociable hours or persistence after refusal).
Tactics used — The specific techniques detected, such as pressure, false urgency or requests for personal data.
Sector — The field the activity belongs to (telecoms, energy, debt collection, surveys, etc.).
Detected company — The company or organisation the number appears to represent or impersonate, when it can be identified.
Impersonation — An indicator of whether the number pretends to be a legitimate entity (bank, public administration, known brand).
Recommended action — The final recommendation for the user: block, caution, ignore or safe.

This analysis is indicative and is generated automatically from the published comments and reports; it does not replace the user's own judgement and is not an accusation against any specific company.

Moderation: what is reviewed and what is not

Reports and comments sent from the NoCall app or website are not published when they are sent: they stay pending until an admin approves them. Only then do they count towards the risk score and become visible to other users. The author can see their own pending contribution; nobody else can. There is no automatic filter for offensive words or personal data: review is manual.

To limit abuse, only one report per number is accepted every 7 days from the same device, IP address or account. In the app, any signed-in user can flag a published comment: a single flag takes it down and sends it back to the review queue.

Number pages also show comments that do not come from NoCall users: text taken from other websites and automatically generated text. No admin has reviewed them, so they do not carry the verified mark and each one states its origin.

Data sources

Each number's assessment combines three independent sources:

Community reports — The foundation of everything. Users report numbers from the website, without an account, or from the app, signed in, choosing a category and adding a comment. Once an admin approves them, they are the main signal in the score.
CNMC operator and prefix data — We use public data from the National Commission for Markets and Competition (CNMC) to identify the operator assigned to each numbering block and the line type (mobile, landline, premium rate).
Company news — We use corporate RSS feeds when available, otherwise Google News feeds.
Comments from other sources — Comments taken from other websites (for example ListaSpam, tellows or unknownphone) and automatically generated comments. They appear on number pages with their origin stated and without review by an admin.