How I AIHow this former NYT columnist uses ChatGPT to brainstorm, do research, and find the perfect metaphor
CHAPTERS
Why Farhad adopted ChatGPT as a writing companion (not a replacement)
Farhad describes his initial skepticism when ChatGPT launched and how rapid improvements turned it into a practical writing aid. He explains how it moved from occasional use to a constant “second window” alongside his draft.
- •Early versions produced weak prose but improved quickly
- •AI felt useful first for small tasks (e.g., finding words)
- •Shift from sporadic consulting to always-on co-pilot workflow
- •Framing: augmentation over replacement for creative work
Brainstorming without Google: using ChatGPT with web search
Farhad shows how he starts idea formation by asking ChatGPT broad, argumentative questions and letting it surface the landscape of commentary. This replaces time-consuming Google searching and manual synthesis at the start of an article.
- •Use web search mode to map the debate around a topic
- •Start with a thesis (e.g., ‘tariffs are good’) to find supporting viewpoints
- •Get a fast overview of key actors, arguments, and narratives
- •Brainstorming becomes interactive questioning rather than link-hunting
Live demo: tariffs query → sources, names, and angles in minutes
In a practical example, Farhad queries commentary on Trump’s tariffs and asks follow-ups (e.g., automotive industry reactions). He emphasizes how quickly the model delivers a structured starting point plus reading links.
- •Prompting for ‘all the commentary’ plus a specific slant
- •Follow-up questions to drill into industries or stakeholder groups
- •Immediate links to articles reduce research ramp-up time
- •Time savings compared with a weekly columnist’s traditional workflow
Evaluating source quality and reducing hallucination risk
Claire probes whether AI-surfaced sources are as reliable as Google. Farhad explains how citations and ‘things it consulted’ improve trust, while still requiring verification.
- •Citations next to claims help validate statements
- •Ability to open the full list of consulted sources
- •News queries tend to surface recognizable outlets (Reuters, etc.)
- •Earlier lack of sourcing made hallucinations a bigger issue
Deeper exploration: AI as a research assistant you can interrogate
Farhad explains that the biggest benefit isn’t just speed—it’s depth. By asking “real questions” in sequence, he can uncover angles and arguments he might not have considered.
- •Helps answer ‘where do I start?’ with suggested main points
- •Can surface overlooked angles and counterarguments
- •Comparable to a research assistant who understands the material
- •Interactive questioning frees time from basic scanning/synthesis
The ‘newsroom in a chat box’: remote collaboration vibes without the social cost
They discuss how ChatGPT mimics newsroom back-and-forth, especially in a post-remote world. Farhad notes the Slack-like interface and the freedom to be blunt without worrying about feelings.
- •Chat interface feels like Slack-based colleague conversations
- •Useful for bouncing ideas when working solo/remote
- •You can say ‘that’s dumb’ and iterate immediately
- •Typos and fast, messy thinking are tolerated—low friction
Finding the perfect idiom/metaphor: replacing clichés like ‘pay the piper’
Farhad demonstrates his signature technique: paste a cliché idiom or rough sentence and ask for fresher, more precise alternatives. This goes beyond what Google or a standard thesaurus can provide.
- •Use AI to search/compare idioms (not just synonyms)
- •Request alternatives that preserve meaning but avoid clichés
- •Discuss nuance, origins, and correctness of idiomatic usage
- •Iterate on suggestions (reject incoherent imagery, refine prompts)
From perfectionism to flow: speeding micro-decisions while keeping authorship
Farhad argues AI reduces time spent stuck on phrasing, letting him keep writing and return to polish later. He emphasizes the result still feels like his voice—AI supplies options, not final authority.
- •Reduces time spent obsessing over one sentence before moving on
- •Supports rough-draft-first workflow with later refinement
- •AI suggestions are ingredients; writer chooses and integrates
- •Addresses fear of ‘slop’ by making writing more specific, not generic
Word-level precision: ChatGPT as a ‘super-thesaurus’ with intent categories
They move from idioms to word choice, using “alternatives to outrage” as an example. Claire notes the tool can organize suggestions by tone (dramatic, colloquial, ironic), helping writers match intent.
- •Faster than searching a thesaurus via Google links
- •Often surfaces broader/more targeted options than a basic thesaurus
- •Can categorize by tone and rhetorical effect
- •Supports rapid exploration of phrasing variations
Checking shades of meaning: ‘Does this word actually fit here?’
Farhad shows how he tests a questionable substitution (e.g., “public grief”) and asks the model to critique it. The AI explains connotation differences and offers rewrites that preserve the desired meaning.
- •Ask directly if a word choice is correct in-context
- •Get connotation/nuance explanations (sorrow vs anger, etc.)
- •Receive alternatives and sentence-level rewrites
- •Use as an always-available semantic check while drafting
AI as a first reader: structural feedback on early paragraphs
Farhad describes feeding 6–7 draft paragraphs to ChatGPT to see if the piece is moving fast enough and whether structure needs tightening. He treats it as iterative feedback while writing, before involving a human editor.
- •Submit partial drafts to assess clarity and pacing
- •Ask for faster paths to the argument and less throat-clearing
- •Useful for phrasing/structure, less for deep logical validation
- •Cycle: draft → AI feedback → revise → write more → polish
Wish list and prompt strategy: memory, screen-sharing, and being blunt
In lightning round, Farhad shares what he wants next (better persistent memory and less copy/paste via screen/app awareness). He also explains his troubleshooting style: direct, sometimes harsh, and knowing when to stop using AI.
- •Pain point: weak persistent memory across chats and time
- •Desire: screen/app integration to avoid constant copy/paste
- •Notes Cursor as an example of tighter tool integration
- •When AI fails: be direct, then disengage if stuck in circles
