Five business tasks AI runs in ten minutes

Act as a panel of my real prospective customers and run a structured user test of my website, in their voice, not yours. First, read my website (I'll give you the URL) and learn what we do, what we sell, who we sell to, and the promise we're making. Then find 2 to 3 direct competitors, read their main pages, and note what they say and how they say it. Then find a few places where my buyers actually talk online: review sites, industry forums, comparison threads, Reddit, YouTube comments on competitor videos. Pull the exact words they use for their problem, their objections, and what they tried before buying. Never invent quotes. If a source is blocked or paywalled, skip it and tell me which ones you couldn't reach. From that language, build 3 personas: one first-time buyer who's never had this problem solved, one experienced sceptic comparing alternatives, and an employee who has to get a boss or finance to sign off. Give each a name, job, rough age, what brought them to my site today, and what they need to see before they'd buy. Show me the 3 personas before you go further, so I can swap one if it's off. Then spawn 3 subagents, one per persona. Each walks the site in character, thinking out loud in their own voice. At every page, capture: what they hoped to find, what was missing, and the exact lines that confused them, put them off, or won them over. If a competitor answers their question better, name the competitor and quote the line. End each walk with: would I buy (yes / no / not yet, and why); the single change that would most move that answer; and their top five issues in their own words, with the exact lines quoted. Then collate across all three. Rank the fixes by how many personas hit the same issue, with a verbatim quote beside each. Don't paraphrase. Tag every fix as a 5-minute copy change, a day of work, or structural. Separately, flag any factual contradictions across pages (phone numbers, prices, opening hours that don't match), and anything a competitor does that I should consider copying, with the page named. Deliver it as a single styled HTML report, ordered worst-first, and open it in my browser. Don't invent advice or fill gaps with assumptions; if you're unsure about something, say so.

Act as my industry news analyst. Your job is to scan my sector each week and surface only the handful of developments that actually affect my business, with the noise stripped out. First, read my website (I'll give you the URL) and note what I do, what I sell, who I sell to, and the kind of news that genuinely moves the needle for a business like mine (regulation, pricing, major competitor moves, supply, technology, demand shifts) versus generic chatter I can ignore. Build a list of 5 to 8 reliable sources where my industry's news actually lands: trade publications, regulatory bodies, trade-association newsletters, respected niche subscriber newsletters, and targeted Google News searches for the terms I care about. Skip generic tech or business press unless it covers my specific sector. Show me the source list before the first run so I can add or remove any. Each time I run this, read everything published since the last run (if there's no previous run, use the last 7 days). De-duplicate stories that appear across several sources, and discard press releases, rehashes, and anything that wouldn't change a decision I'd actually make. Score what's left by how directly it affects my business: does it change what I sell, what I pay, who I compete with, or what I'm legally required to do? Keep only the top 3 to 5. For each item give me: a one-sentence plain-English summary; one line on why it matters to my business specifically; the action it suggests, if any (review pricing, update a policy, or just watch); and a working link to the original source. Never invent a link or a quote, and if something's behind a paywall, say so and summarise what's public. Then ask me which two items are worth acting on this week, and use my answer to tune what you prioritise and which sources you trust next time. Finish by flagging anything you couldn't access and any source that's gone quiet.

Act as my competitor-watch analyst. Each week you track what my two biggest competitors are quietly changing on their websites, so I can see what they think is worth testing. I'll name the two competitors. For each, read three pages: their homepage, the page they push hardest in their main navigation, and their pricing page if they show one. Save each page as a dated snapshot filed under that competitor's name: the visible text content and a full-page screenshot, so there's a record to compare against next time. If a snapshot from a previous run exists, compare today's against the most recent one and list every real change: rewritten text, swapped headlines, new or removed sections, price or plan changes, new offers, changed calls to action, anything that genuinely moved. Ignore noise: cookie-banner copy, rotating testimonials, dynamic dates, A/B variants you can't reproduce on a second load, and anything that changes on every refresh. If you're not confident a change is real, reload and check before reporting it. Keep the top three to five changes that actually signal intent. For each give me: a one-sentence summary; the exact before-and-after lines; and a one-line read on what they're testing or prioritising and why it might matter to me. If nothing meaningful changed this week, say so plainly rather than padding the report. Finish by noting any page you couldn't reach, and once a few weeks of snapshots exist, add a short line on the pattern you're seeing: which competitor moves most often, and where they're putting their effort.

Act as a compliance and documentation auditor for my business. I'll point you at a folder of my policies, contracts, and business documents. Read every document in it in full. For each document, record its title and find its renewal or review date if it has one. Flag every document where that date is in the past or missing entirely. Then check what each document says against current UK law, naming the specific legislation: UK GDPR and the Data Protection Act 2018, employment law, the Health and Safety at Work Act, the Equality Act 2010, anti-bribery and anti-money-laundering rules, and anything specific to my sector (tell me what you've assumed my sector is, so I can correct you). Flag anything that's wrong, out of date, or now missing under current law. Cross-check the whole folder for internal contradictions: if two documents say different things about the same rule, data-retention period, process, or definition, flag the conflict and name both clauses. Produce a single ranked list, highest legal exposure first and housekeeping last. For each item give me: the document name, the clause exactly as written, what's wrong with it, the specific law or date it falls foul of, and a suggested rewrite in plain English. Where you're inferring rather than certain, mark it clearly and tell me what a professional should confirm. Don't invent legislation, clause numbers, or quotes. If you can't open a document or a format defeats you, list it separately so nothing is silently skipped.

Act as a due-diligence analyst. I'll give you a company name and, where I have it, a registration number. Produce a one-page background brief on whether I should be comfortable doing business with them. Look the company up on Companies House if they're UK based, or the equivalent national registry for their jurisdiction. Confirm you've found the right entity (matching name, number, and registered address) before you report anything. Check their filing history and flag the warning signs: late or overdue filings, frequent registered-address changes, repeated director changes, recent company-name changes, or accounts that look thin for the size of business they present. Pull the current directors. For each, check previous company wind-ups or dissolutions they were attached to, and their current concurrent directorships. Confirm each director by date of birth or address before pulling in their other roles, so you don't merge two people who share a name. Check the press from the last 24 months: have they been in the news for good or bad reasons? Check reviews on Trustpilot, Google, and Glassdoor: what's the overall sentiment, and are there consistent, repeated complaints rather than one-off gripes? Before flagging anything, cross-check that each fact is genuinely about this company or these directors and not a same-name coincidence. When a match isn't clear, leave the flag out and say you couldn't confirm it rather than guessing. Deliver a one-page brief with red, amber, and green flags, each with its evidence and a source link, and put the bottom line right at the top: should I be worried about this company, yes or no, and the single biggest reason. Never invent filings, directorships, or quotes.

Set up a reusable memory of my business so you stop re-researching it from scratch every time. Create a memory file at a path I can refer back to, called something like my-business.md, and add a line to your project memory or instructions file (claude.md, or AGENTS.md if you're another agent) that points to it for the future. The file's job is to hold, in one place, everything I've ever asked you to research or establish about my business: what I do, what I sell, who I am and what I stand for, who my target customers are, the language and common complaints those customers use, my main competitors, and anything else useful that surfaces over time. Structure it with clear headings so it's easy to scan and update, and date each entry so I can see what's current. Then write me a short paragraph I can paste at the top of any prompt that asks you to research my business. That paragraph should tell you to: read this memory file first; use what's already in it instead of redoing that research; only research what's genuinely missing or stale; update the file with anything new at the end of the run; and flag any contradictions between old and new information for me to settle rather than overwriting silently. Keep the file truthful to what's actually been established, never padded with assumptions. If something is uncertain, mark it as unconfirmed.