The podcast that was too hard and the coursebook that was too easy

Separating these matters, because each has a different remedy.

Text features that were genuinely hard:

• Length: four minutes of continuous speech demands sustained attention far beyond what the class has practised.
• Natural speed and connected speech: weak forms, elision, assimilation - the learners have been listening to scripted, clearly articulated recordings.
• Two unfamiliar voices with overlapping turns, plus one regional accent.
• Idiomatic and low-frequency lexis they had no way to infer quickly.

Task features that I got wrong:

• Five questions on four minutes is a very high information density - I was asking for near-complete comprehension. Detail questions also asked them to catch specific words, which is precisely what connected speech destroys.
• No pre-listening: no orientation to topic, genre or speaker relationship, so they had no schema to predict from.
• One listen to a hard text, then comprehension questions - no staged listening.
• Learners who finished early had nothing else to do, which is a task-design fault, not a learner fault.

Learner factors:

• They are decoding word by word rather than chunking; "too fast" is usually a decoding-speed problem, not a tape-speed problem.
• They treat the transcript as the answer rather than as a post-task support - a habit I have probably created by using transcripts to "rescue" listening lessons.

So the main fault is mine: I graded the text instead of grading the task.

The text stays because it is the right level of challenge if the task is scaffolded. Sequence:

Before listening (7 min): Show who the speakers are and their relationship; one sentence on the topic. Learners predict three things the speakers might disagree about and compare in pairs. Pre-teach five lexical items that are genuinely unguessable and useful - not every unfamiliar word.

First listen - gist only (4 min): One question: What are they arguing about, and who do you think is right? No detail. Play the first 60–90 seconds only.

Second listen - narrow task (6 min): Three items, not five, focused on the argument structure rather than on exact words. Pause at two marked points to let learners compare in pairs - this converts a memory test into a listening task.

Third listen - chunked (6 min): Play 30–40 second sections with a specific micro-task for each (e.g. Does she agree or disagree here?). Early finishers get a second question on the same chunk, not a new chunk.

After listening (8 min): Transcript revealed only now, for noticing: learners find two places where words ran together and mark the weak forms. This turns the transcript into a decoding lesson rather than a crutch.

Recycling: Same extract replayed in the next lesson as a warmer with no transcript, plus a two-minute "fast speech" slot each week with short authentic clips so speed is normalised incrementally.

I would not skip it, and I would not pretend it is a listening comprehension lesson. It is a resource with a different job.

• Use it for strategy and decoding. A short dictogloss: play twice, learners reconstruct in pairs, then compare with the script. That trains noticing of form and connected speech, which is exactly what the class lacks.
• Use it as a model for pronunciation and functional language. The script is clean, so it is ideal for marking stress, intonation and the exponents of the function it contains, then transferring them into a role-play.
• Use it for prediction-checking or as a lead-in to the harder authentic text rather than as the main listening.
• Build a progression ladder for the term, so difficulty rises deliberately rather than by accident: length 50 sec → 1 min → 2 min → 4 min; scripted → semi-scripted → authentic; one speaker → two; with transcript support → without. The authentic podcast was not wrong, it was four rungs up the ladder.

The judgement I am applying: difficulty in listening is a property of text and task and learner skill. When a text is easy, I raise the task demand; when a text is hard, I lower the task demand and add scaffolding. Neither response is "find a different recording".