Claude Fable 5 Prompt Hacks: Get Real Work Out of Anthropic's New AI
Anthropic's new flagship reads a million tokens and checks its own work — but old prompting habits make it worse. The rules changed; here's the new playbook.

Claude Fable 5 is Anthropic's new flagship — a "Mythos-class" model that sits a full tier above last year's Claude Opus, reads up to a million tokens of context (roughly 700,000 words), and can run multi-day projects with minimal hand-holding. It relaunched broadly in July 2026, and small business owners are quietly discovering it's the best back-office employee $20 a month ever bought. But here's the catch: most people prompt it like it's 2024, and old habits actively make Fable 5 worse. Here's how to get your money's worth.
The one rule that changes everything: give it goals, not steps
Anthropic's own guidance for Fable 5 is blunt: stop writing step-by-step instructions. Earlier models needed micromanaging; Fable 5 performs worse when you over-specify, because your mediocre plan overrides its better one. Describe the outcome and how it should know it's done — then get out of the way.
Old habit: "Write a headline. Then write three paragraphs. Paragraph one should say... Use bullet points for..."
Fable 5 habit: "I need a services page that makes a stressed homeowner call us instead of the two bigger companies in town. It's done when a stranger could tell in 10 seconds what we do, where, and why us. Here's everything about my business: [paste it all]."
Hack #2: one rich brief beats twenty little messages
Fable 5 thinks before every reply — that's why its answers are strong, but it also means each round-trip costs you more waiting than older models. The efficient pattern is fewer, fatter turns: front-load everything into one message instead of drip-feeding across a conversation. With a million tokens of context, "everything" is meant literally — your whole price list, your last 50 reviews, three competitor websites, that rambling voicemail transcript. It will use all of it.
Hack #3: make it check its own work
Fable 5's signature strength is self-verification — in coding harnesses it runs its own tests, and you can trigger the same behavior in plain chat by adding one line to any prompt: "Before you answer, verify every claim against the facts I gave you, and flag anything you had to assume." That single sentence catches most of the invented-detail problems that make business owners distrust AI writing.
Five copy-paste prompts that earn their keep
The review-reply batch. "Here are this month's reviews [paste]. Draft replies in my voice: grateful but not gushing, specific to what each customer mentioned, 2–3 sentences. For anything under 4 stars, acknowledge, take it offline, no excuses. Flag any reply I should legally not send."
The quote decoder. "I got this estimate from a contractor/vendor [paste]. Explain it line by line like I'm smart but not in this industry. What's missing, what's padded, and what three questions should I ask before signing?"
The job-story multiplier. "Here's a job we finished this week [describe + paste any photos]. Turn it into: a Google Business Profile post, two Facebook posts, and a 150-word case note for my website. Same facts, no invented details, US spelling, plain language."
The no-show killer. "Write a 3-message reminder sequence (email + two texts) for booked appointments. Friendly, zero guilt-tripping, each shorter than the last. My cancellation policy: [paste]."
The Monday planner. "Here's my unsorted week: [paste calendar, to-dos, that email thread]. Build my week: what to do, what to delegate, what to ignore, and the one thing that makes everything else easier if done first. Be opinionated."
Hack #4: paste, don't describe
The most underused Fable 5 ability is reading documents — it genuinely understands charts, tables, and diagrams buried inside PDFs. Don't summarize your P&L or that 30-page insurance policy for it; attach the actual file and ask your question. Summarizing first throws away exactly the details it's better than you at noticing.
Hack #5: know when you're over-buying
Honesty corner: Fable 5 is premium horsepower, and plenty of daily tasks don't need it. A quick email subject line or a reworded sentence runs fine on lighter, cheaper models — Anthropic itself positions Fable 5 for complex, multi-step work. The rule of thumb we give clients: if the task involves judgment across lots of information (analyze, plan, audit, negotiate), reach for Fable 5; if it's a one-liner, anything works. Developers get an explicit "effort" dial on the API; in the chat apps, your effort dial is simply how much real context you feed it.
The fine print for your website
Everything Fable 5 drafts for your website still needs your facts and your edit before publishing — not because Google penalizes AI writing (it doesn't; it penalizes lazy publishing), but because unedited AI content is exactly what the June 2026 spam update learned to smell. Use the verification line from Hack #3, add the one detail only you could know, and you're on the right side of both Google and your customers.
And while you're thinking about AI: your customers' assistants — including Claude itself — are already deciding whether to recommend your business. Check whether AI can even read your website with our free AI Visibility Checker, or start with how to get recommended by ChatGPT.