Scripted vs Unscripted Calls: Why Speaking Tests Matter More Than Reading

The gap between reading a script and holding a conversation

A candidate can read a rebuttal script flawlessly and still freeze the moment a customer goes off-book. That gap is the single most expensive blind spot in phone-based hiring. Reading a prepared line is a recognition task performed at the reader's own pace. A live, unscripted call is a production task performed at the customer's pace, in the customer's accent, with the customer's interruptions. They draw on different abilities, and an assessment test that only measures the first will quietly over-promise on the second.

This matters because most call-centre work is unscripted in the parts that decide the outcome. The greeting and the compliance disclosure may be scripted, but the objection, the confused question, the frustrated escalation and the empathetic recovery are not. If your screen rewards script reading and grammar recognition, you will hire people who look strong on paper and churn on the floor. Below we break down what unscripted calls actually demand, why reading-heavy tests mislead, and how to design a screen that samples real speech.

What an unscripted call actually demands

Strip a live call down to its component skills and almost none of them are exercised by a silent multiple-choice quiz. A capable phone agent has to do all of the following, in real time, at once:

• Comprehend varied accents at speed. Customers do not enunciate for the benefit of a non-native listener. Regional accents, background noise and fast, colloquial speech all land in the same second the agent has to respond to them.
• Produce speech spontaneously. There is no answer bank to select from. The agent has to assemble a grammatical, intelligible sentence on the spot and keep talking while thinking.
• Handle interruptions and objections. Customers talk over the agent, change the subject and push back. The agent has to track the thread, hold their point and stay coherent when the call stops following any predictable path.
• Carry empathy under pressure. Tone, pacing and word choice have to convey calm and care even when the customer is angry and the handle-time clock is running.
• Recover from their own errors. When the agent mishears or misspeaks, they have to repair the sentence mid-flow without losing the customer's confidence.

None of these are visible in a reading comprehension score. A candidate can have an excellent passive vocabulary and still be unable to deploy it fast enough to survive a hostile call. Deciding what English level a call center agent needs starts with recognising that the job is a speaking-and-listening job first, and a reading job a distant second.

Why reading-heavy and grammar-only tests over-predict

Reading and grammar tests are popular for practical reasons. They are cheap, they auto-mark instantly, and they produce a tidy percentage. The problem is not that they measure nothing — it is that they measure the wrong thing for this role and then present a confident number that hiring managers trust.

Three mechanics drive the over-prediction:

1. Recognition is easier than production. Choosing the correct option from four is a lower-effort task than generating the correct sentence from nothing. A candidate's recognition ceiling sits well above their production ceiling, so a multiple-choice score systematically flatters spoken ability.
2. Self-paced tests hide processing speed. On a reading test the candidate controls the clock. On a call the customer controls it. A test that removes time pressure removes the exact variable that breaks weaker speakers.
3. Grammar knowledge is not fluency. Knowing that a sentence is wrong is not the same as building a right one under load. Plenty of candidates can spot a tense error on paper and still stall when they have to speak.

The result is a comfortable-looking pass rate that does not survive contact with the phone. You interview a shortlist that reads as fluent, then discover in the first week of nesting that a chunk of them cannot sustain a real conversation. The cost lands as early attrition, re-hiring and lost ramp time. We looked at whether English testing reduces agent attrition and the pattern is consistent: screens that sample the real skill retain better than screens that sample a proxy for it.

Passive MCQ versus a test that captures speech

The contrast is easiest to see side by side.

How to design a screen that samples spontaneous speech

Weighting speaking more heavily does not mean dropping the other skills. It means matching the weighting to the job and insisting that the primary gate produces actual speech. A screen that works for voice roles tends to share a few features.

Sample all four skills, but weight them for the role. Listening and speaking should carry the decision for a phone job; reading and writing become supporting evidence for CRM notes, email and chat handling. A test that reports a per-skill breakdown against the CEFR framework lets you set that weighting on purpose rather than hoping a single blended number reflects it.

Use open-ended speaking prompts, not read-aloud tasks. Asking a candidate to read a paragraph tests pronunciation, not conversation. Asking them to answer an unpredictable question forces the spontaneous production that a real call demands. The prompt should require the candidate to generate their own content, because that is the ability you are hiring for.

Build accent variety into the listening items. If your customers speak in a range of accents, your listening section should too. A candidate who only ever practised on clean, slow audio has not been tested against the acoustic reality of the floor.

Score consistently and fast enough to use. Human panels are accurate but slow and hard to standardise across a high volume of applicants. AI-scored speaking closes that gap: it captures the recorded response and returns a CEFR level per candidate in minutes, which means you can sample spontaneous speech for every applicant instead of only the shortlist that already passed a cheaper proxy. International English Test scores all four skills — listening, reading, speaking and writing — with automated and AI scoring, with speaking assessed by AI and a CEFR level (A1 to C2) returned per candidate quickly, on credit-based pricing of roughly £8.99 to £11.99 per test by volume and no contracts. As an ALTE Associate Member, it maps results to the same CEFR standard your other language evidence uses.

Set a defensible threshold and apply it uniformly. Once you can see spoken CEFR per candidate, pick the minimum that the role genuinely needs and hold the line. A consistent, speech-based bar is fairer to candidates and more predictive for you than an interview panel's gut read after a fluent-looking paper score.

Position the screen before the interview, not after. The whole point of a speech-based gate is to protect scarce recruiter and team-leader time. If you interview first and test later, you have already spent the expensive resource on candidates who a two-minute speaking sample would have filtered out. Put the speaking screen at the top of the funnel so that everyone who reaches a human conversation has already demonstrated they can sustain one. This also standardises the first impression: instead of comparing candidates against different interviewers on different days, you compare them against the same CEFR-scored prompts.

The bottom line for phone-based hiring

The ability to read a script tells you a candidate can recognise English. It tells you very little about whether they can hold a live, unpredictable conversation with a real customer who is not reading from anything. For voice roles the two are not interchangeable, and an assessment test that leans on reading and grammar will keep over-predicting on the skill that actually decides the call.

Weight the screen the way the job is weighted: speaking and listening first, reading and writing as support. Sample real, spontaneous speech from every applicant, score it consistently against CEFR, and set a threshold you can defend. If you are building that screen now, start with an English test built for call center hiring that samples the skills the role really uses.

English assessment tests for companies