PTE Academic is scored entirely by software, on the 10–90 Global Scale of English, and that fact should change how you prepare rather than just how you feel about it.
Integrated scoring. Most tasks contribute to more than one skill. "Read aloud" feeds both reading and speaking. "Summarise written text" feeds reading and writing. "Write from dictation" feeds listening and writing. So a weak area quietly drags two scores down, and the highest-value tasks to drill are the ones that count twice.
The machine wants fluent, continuous speech. In the speaking tasks, long pauses, restarts and "um" cost you directly — the algorithm measures oral fluency as its own trait. It is better to say a slightly wrong word smoothly than to stop and repair. This is the opposite instinct from a human-marked exam, and it is the single biggest adjustment for students coming from IELTS.
Content is scored literally. In "Describe image" and "Retell lecture", the software is looking for the key words that belong to that content. Say the numbers, the labels, the trend words. A beautifully delivered answer that never names what is on the screen scores badly.
Write from dictation is worth a lot for a short task. Every correctly spelled word counts. Many candidates gain several points just by drilling this one item type.
The timing is fixed and unforgiving. Each item has its own countdown and there is no going back — PTE does not let you revisit a section, so a question you are unsure about is answered and left.
Two practical habits: practise speaking into a microphone rather than to a person, because recording changes everyone's rhythm; and learn the item list itself — there are about twenty item types, and knowing which one is coming is half the preparation.