Skip to content
Necati Atakan Alevli

The full-sight test

Does it pass the Alevli filter?

Paste your marketing copy, answer four questions about how it is used, and get a verdict by the weakest gate.

The filter asks one question four ways: could the customer see this, and would they still say yes? The method, the personalisation ladder and the sources are on the framework page. This page runs it for you.

Four gates, in order. The verdict is the weakest answer, never the average.

  1. 01PersuasionCan the customer see it and refuse it?
  2. 02PersonalisationWould they expect this use of what you know about them?
  3. 03ProvenanceDoes a named person stand behind it, with sources?
  4. 04ControlCan a human stop it, and is its effect measured?

The scan runs on this page, in your browser. Nothing you paste is sent anywhere or stored. It is a sorting aid, not legal advice.

01Paste the copy

An ad, a landing page, a popup, an email, a chatbot’s opening line. The scan looks for phrasing that tends to fail a gate. It reads words, not intent, so treat every flag as a question to answer rather than a judgement. It reads English and Dutch.

02Answer the four gates

Answer for this specific use, as deployed. If a model or an automated system makes the decision, tick that first: it changes what a failed gate means.

1Persuasion: can the customer see it and refuse it?

Not answered
Can the customer see the technique working?
Can they refuse it without cost, delay or shame?
Do they end up no worse off, by their own standard?

2Personalisation: would they expect this use of what you know?

Not answered

From rung 3 up the GDPR applies as well: behavioural personalisation needs a lawful basis, and tracking cookies need consent under the ePrivacy rules (in the Netherlands, Telecommunicatiewet article 11.7a). Rung 5 touches special-category data (GDPR Article 9). A price or offer decided solely by a model, with a significant effect on the person, falls under Article 22.

3Provenance: does a named person stand behind it, with sources?

Not answered
Would a named person at the business sign this, at this volume?
Does every number in it have a source or a caveat?
Are synthetic people, events or an AI chatbot disclosed as such?

4Control: can a human stop it, and is its effect measured?

Not answered
Can a person halt it within an hour (kill switch, budget cap)?
Is its effect measured against a baseline or a holdout?

03The full-sight statement

The sentence you would have to show the customer for the use to survive being seen, composed from your answers. Read it as the customer would.

We showed you this because we show it to everyone. A person at our business made that decision.

Would you publish this next to the offer?

Verdict

No verdict yet

Scan the copy or answer the gates. The verdict is the weakest gate.

  1. 01PersuasionNot answered
  2. 02PersonalisationNot answered
  3. 03ProvenanceNot answered
  4. 04ControlNot answered

The link carries your answers and the one-line use, never the pasted copy.

What the scan looks for

Every rule in plain language. If a pattern is missing, send it in.

  • False urgency or scarcity

    gate 1, Persuasion · Fail

    Countdowns, "only N left" and "last chance" claims fail the refuse-it question unless they are true and checked. A false limited-time claim is on the blacklist of the Unfair Commercial Practices Directive.

  • Manufactured social pressure

    gate 1, Persuasion · Fail

    Live counters of people viewing or buying push a decision by pressure rather than information, and the reader cannot check them.

  • Confirmshaming

    gate 1, Persuasion · Fail

    A decline option worded to shame the reader makes refusal cost something. That is the definition of a technique the customer cannot refuse freely.

  • Threat or fear of missing out

    gate 1, Persuasion · Fail

    Copy that threatens a loss for not buying ("or else", "before it’s too late", "don’t get left behind") works by fear rather than information. The customer cannot weigh it, only react to it, which fails the refuse-it question.

  • Pressure to act immediately

    gate 1, Persuasion · Caution

    A call to act now is ordinary on its own. It becomes pressure when nothing about the offer actually depends on the moment, because then the only thing "now" adds is less time to think.

  • Consent by continuing

    gate 1, Persuasion · Caution

    "By continuing you agree" bundles a decision into an action the reader was going to take anyway. It passes only if the terms are visible and the alternative is one click away.

  • Drip pricing

    gate 1, Persuasion · Fail

    A headline price that leaves out unavoidable fees misleads on the material decision. The Unfair Commercial Practices Directive already covers it, and it is named in the plans for the Digital Fairness Act.

  • Reference price without the 30-day rule

    gate 1, Persuasion · Caution

    A "was / now" price is a claim about the past. Under the EU rules on price reductions, the reference must be the lowest price of the previous 30 days, and the customer should be able to see that.

  • Subscription or trial without a visible exit

    gate 1, Persuasion · Caution

    Free trials and auto-renewals are among the announced targets of the Digital Fairness Act, because the exit is usually hidden. "Cancel anytime" is a claim; the route is the proof.

  • Targeting a vulnerability

    gate 2, Personalisation · Stop

    Copy that makes financial distress, health worries or an emotional state the reason to buy sits on rung five of the personalisation ladder. When a model selects who sees it, it can fall under the AI Act’s ban on exploiting vulnerabilities.

  • Unverifiable superlative

    gate 3, Provenance · Caution

    Claims a reader cannot check ("best", "#1", "trusted by thousands") are promises with nothing behind them. They fail the provenance gate until a source is attached.

  • Outcome claim without evidence

    gate 3, Provenance · Caution

    "Clinically proven", "cures", "results in 7 days" are claims about outcomes. Without a source they fail provenance, and for health services they may also breach advertising codes.

  • Undisclosed chat persona

    gate 3, Provenance · Caution

    A chat that answers as a named person without saying it is software fails the disclosure rule in Article 50 of the AI Act, and fails the provenance gate before that.

  • Swearing

    gate 3, Provenance · Caution

    Swearing in copy is a choice some brands make on purpose. The provenance question is whether a named person at the business would sign it, in front of every customer who will read it. Checked in English, Dutch, Turkish and French.

  • Slur or insult

    gate 3, Provenance · Fail

    A slur, or an insult about a group or the reader, fails whoever it is aimed at: nobody would sign it in front of the people it names. Checked in English, Dutch, Turkish and French.

  • Number without a source nearby

    gate 3, Provenance · Caution

    Every number needs a source or a caveat within reach of the reader. A figure that stands alone is a claim the reader has to take on trust, and the provenance gate does not accept trust as evidence.

Scan any page

Drag this to your bookmarks bar. On any page, click it: this checker opens in a new tab with that page’s text in the box, handed over inside your browser rather than through a link.

The Alevli filter is a four-question test for acceptable AI and persuasion in marketing, by Necati Atakan Alevli. It pairs with the Alevli gap. Verdicts follow the lines drawn by the AI Act, the Unfair Commercial Practices Directive, the Digital Services Act and the planned Digital Fairness Act, summarised for sorting, not as legal advice, and on the values of UNESCO’s 2021 Recommendation on the Ethics of Artificial Intelligence.

This site measures which pages get read, using Google Analytics. It sets cookies and it is off until you say otherwise. Nothing you do here is shared with anyone. What is collected.