The Virtue of Persistent Effort (Iteration)

Perplexity does not iterate on prose - it has no document surface, and it will cite sources inside anything it writes. But almost every draft that needs a second pass needs it because the first was built on assumptions, and that is exactly the gap Perplexity fills: do the grounding round here, with sources and dates, then take the findings into a tool that will draft and revise. Spaces keep the research alongside the follow-up questions.

• Use it for the grounding round, not the drafting rounds.
• Keep a thread's research in one Space for follow-ups.
• Ask it to check your simplification is not actually wrong.
• Academic focus for research claims, Web for prices and news.
• Ask it to flag single-source claims rather than smoothing them.

Perplexity will not write your noir story well - it searches, cites and summarises, and a cited short story is not a short story. But iteration has a research step, and this is the step it is good at: the opening draft of invented-city fiction almost always fails because the city is generic, and the fix is real detail.

I am writing noir fiction set in a rain-soaked Neo-Hyderabad in 2077. Research this with sources:

1. What are the defining physical and sensory details of Hyderabad that a novelist would use - the old city and Charminar area, the IT corridor, the rock formations, what the monsoon actually does to the streets? Cite travel writing, reportage and local accounts rather than listicles. I want specifics I could not have guessed.
2. What are Hyderabad's working markets like - which ones, what trades, what hours, what they sound and smell like? Cite accounts with dates.
3. What is on the record about Hyderabad's planned development, infrastructure and climate projections that a 2077 setting should extrapolate from rather than invent?
4. Which Indian noir and Indian science fiction writers have done city-as-character well, and what do critics say they did differently? Cite the criticism.
5. What are the named cliches of cyberpunk city writing that reviewers consistently call out - neon, rain, noodle stalls? Cite the critiques so I know what to avoid.

Cite everything, with dates. Save it to a Space called "Neo-Hyderabad research".

Then I will take this into ChatGPT or Claude and iterate on the actual prose there, with these details as the raw material. That is the honest division of labour: research here, drafting there.

This is the one example in this chapter that Perplexity is genuinely better at than the others - not because it iterates, but because the first iteration of a marketing plan usually fails on facts, and this is where you get facts with sources attached.

I run an independent coffee shop in Kondapur, Hyderabad, selling mostly to IT professionals from nearby offices. Marketing budget ₹20,000 per month. Before I write a plan, research the ground it has to stand on:

1. Which cafes and coffee chains currently operate within about two kilometres of Kondapur, and what are they charging and promoting? Cite listings and recent reviews with dates.
2. What are current Swiggy and Zomato commission rates, promotion structures and minimum discount requirements for a small restaurant partner in India? Cite the most recent published figures and flag anything you can only source to one place.
3. What do Hyderabad IT-corridor office timings and hybrid-work patterns look like now, as reported? This decides my peak windows and whether a lunchtime offer makes sense at all.
4. What does published work on independent cafes say actually drives repeat visits - loyalty schemes, seating and wifi for laptop workers, subscription models? Use Academic focus for this one. Cite the studies and say which were done on Indian markets rather than Western ones.
5. What has been reported on realistic returns from ₹20,000-a-month local marketing for a single-outlet food business in an Indian metro? If nobody publishes credible numbers at this scale, say so plainly rather than estimating.

Save all of it to a Space called "Deccan Brews research" so follow-up questions keep the context.

Then take the findings into a writing tool and iterate on the 4-week plan there.

Iterating on the wording of an explanation is not a Perplexity task - it will search and cite mid-sentence, and the cited version is not the explainer you want. But it is the right tool for one iteration step the others cannot do: checking that your simplification is not quietly wrong.

I am explaining the attention mechanism in a transformer model to business managers with no technical background, using this analogy: a student deciding which sentences in a textbook matter for answering a specific question, with the model's scores on words paralleling the student's choices.

Research this:

1. What do the original transformer paper and subsequent explanatory literature actually say attention does? Cite the primary sources. Use Academic focus.
2. Where does the student-highlighting-a-textbook analogy break down against that description? Name the specific mismatches and cite what you are comparing against. This is the main thing I need.
3. Which analogies for attention are in widest circulation in non-technical explainers? Cite examples - I want to know which ones my audience may already have heard, and which ones specialists have publicly criticised as misleading.
4. Has anyone written about the cost of over-simplifying attention for business audiences - decisions people make on a wrong mental model? Cite it if so.
5. What has changed in how attention is explained or implemented that a presentation given now should not get wrong? Cite with dates.

Cite everything. Then I will take the list of mismatches from point 2 into another tool and iterate the wording there, with your corrections as the constraint - which is more useful than iterating on prose that sounds good and is subtly false.