Source: raw/How_I_Use_Claude_Cowork_to_Get_Instant_Press_Coverage.md, raw/How_To_Scrape_Any_Website_With_Claude_Cowork_For_Free.md (auto-caption transcripts, fetched 2026-10-27)
Creator: Eliot Prince Platform: YouTube Videos:
- “How I Use Claude Cowork to Get Instant Press Coverage” — https://www.youtube.com/watch?v=iNpPj-7VH4w — 22:21 — published 2026-04-15
- “How To Scrape Any Website With Claude Cowork For Free” — https://www.youtube.com/watch?v=C_lWVKkYz2k — 13:44 — published 2026-03-16
Eliot Prince, a former SEO-agency specialist, ran a full digital-PR campaign inside Claude Cowork: pick a counterintuitive question, scrape about 1,000 UK job ads through the Apify connector, analyse them, and turn the findings into a branded report, a press release and a data page on his own site. He reports coverage in Business Insider, MarketWatch and Yahoo Finance and 381 other placements. This article lays out the pipeline stage by stage, with his prompts and his cost figures. The earlier scraping video supplies the connector setup and the first small test. Both videos predate the September 2026 chat/Cowork merge, and the press video uses Opus 4.6; the pipeline depends on neither.
Key Takeaways
- The product is a proprietary number. Eliot’s rule: “the best PR studies sit between the gap between what everyone else assumes and what the data actually shows.” His angle was the media story that AI is taking jobs versus how few UK job ads ask for AI skills at all.
- Business context goes in before ideation. With his business DNA files and memory loaded, a short prompt returned 10 study ideas fitted to his AI coaching business. Without that context he expects weak ideas.
- Ask Cowork for a plan before it collects anything. He had Cowork propose roles, sample size (1,000-1,500 job descriptions) and a collection method before running it. Its recommended option was the Apify connector.
- Scraping costs very little. On Apify’s free 1.15 of spend in a month, 5 cents for 50 LinkedIn jobs, 50 cents for 500, and about $1 for the 1,000-job run. These are his figures from his own account.
- Collect, then analyse in separate steps. First pull everything into a spreadsheet as the reference copy; then analyse it under a “data journalist” role prompt. The 1,000-job scrape returned 956 unique rows after Cowork removed 44 duplicates, in about 5 minutes.
- Fact-check before you publish. His sequence includes an “objectively fact-check this” step, which he skipped on camera. His own runs gave different headline numbers (see Results), so this step matters.
- Publish a data page, then distribute. A page on your site holding the findings, charts and raw data gives journalists something to cite and link to. He distributes the press release with Press Ranger (media database plus wire distribution) instead of emailing journalists from Gmail.
The pipeline
1. Find the angle
- Prompt Cowork to create “a data study that [gets] picked up by journalists” for your business, with your knowledge base in context (“You know all of my business and my target audience…”).
- His shortlist included hours per week knowledge workers spend on AI, an “AI confidence gap”, the hourly rate of a consultant with and without AI, and the 2026 state of AI adoption in UK small businesses.
- Pick an idea that data you can actually collect will settle. His was a UK AI jobs report: do day-to-day, admin-heavy job descriptions (personal assistants, receptionists, bookkeepers, accountants) mention AI tools (Claude, ChatGPT, Copilot, Perplexity) or AI terms?
2. Plan the collection
- Ask for a project plan with options before Cowork does anything. Eliot says that if he had just said “go”, Cowork would probably have got stuck or done things in an order that made no sense.
- His plan fixed role categories (roughly 100 jobs per category: admin, bookkeeping, marketing, customer service, sales, social media and others), a 1,000-1,500 sample, and Apify as collection option A.
3. Connect Apify (and optionally Vibe Prospecting)
- Apify: + → Connectors → Manage connectors → Browse connectors → search “Apify” → install → configure. Create an API token in the Apify console (Settings → API and integrations, or Create → new token, where you can set an expiry and limits) and paste it into the connector. Keep the token private.
- Apify is a marketplace of ready-made scrapers (“actors”) for Google Maps, LinkedIn, TikTok, YouTube, Google search, Instagram, Reddit, Facebook ads and general website crawls. You don’t need to name the actor: Cowork searches Apify and picks one. In the March test it picked a Google Maps scraper for local businesses and a LinkedIn jobs actor it described as having 26,000 users and a 4.9 rating.
- Cowork asks permission before calling an actor (“allow once” or “always allow”).
- Vibe Prospecting (for lead lists rather than PR) installs as a connector with an OAuth authorisation and no API key. Eliot reports 400 free credits on a free account. It adds decision-makers’ names, titles, emails and LinkedIn profiles to a company list Cowork has already scraped.
4. Scrape into a spreadsheet
- His prompt (WhisperFlow dictation, lightly cleaned): “I have set up Apify so that you can access that connector and use the LinkedIn scraper to go and scrape 1,000 LinkedIn job descriptions from those categories… Please bring all of that data and place it inside a spreadsheet for me so that we always have it as reference before we actually start the analysis.”
- Cowork ran 10 batches in parallel through Apify. Output columns: role category, job title, location, full job description, employment type.
- Start small. The March test scraped 50 UK administrative listings in about 3 minutes before any large run. The local-business test (20 solar installers in south-west England) returned no email addresses; the Vibe Prospecting step supplied them.
5. Analyse as a data journalist
- His rewritten prompt opens “You’re a data journalist, and your expertise is turning raw data sets into clear, compelling narratives”, then states the task and output format. He switched to Opus 4.6 for this step; on current plans that would be Opus 5.5.
- Ask for the surprising numbers, the quotable statistic, and which roles lead and which lag.
- Then run a separate fact-check prompt over the findings.
6. Package the findings
- Report: a prompt asking for headline findings, detailed breakdowns, methodology and the raw data, with rules. Cowork produced a branded PDF with charts. Load any tone-of-voice skills at this step.
- Press release: “only use the information from the study, don’t use any hype”, in a set format (key findings paragraph, about section, contact details, link to the data page). Eliot would add a 300-400-word cap and write his own quotes rather than keep the generated ones, which he says didn’t match his voice.
- Headline: the first draft (“Fewer than one in 10 UK job ads name an AI tool”) was weaker than he wanted; he suggests leading with the strongest number or with what it means for UK AI adoption.
- Data page: build a page in your brand (in Claude, or a site builder such as Replit, Lovable or WordPress) with the story, interactive charts and downloadable raw data, so coverage links back to you.
7. Distribute
- Press Ranger: publish the release, search its media database for journalists, podcasters and publishers, and send to wires. Eliot says the whole process from ideation to press release took 1 hour 2 minutes of recording time.
Results (creator-reported)
- Coverage: Business Insider, MarketWatch, Yahoo Finance, plus “381 other placements around the web” by his count, and a data page on his site positioning him on AI skills for businesses. None of this has been checked independently here.
- Cost comparison: he puts an agency PR retainer at $5,000 a month or more with no guaranteed results.
- The headline numbers move between runs. The April 1,000-job run gave 0.9% of everyday-role ads naming an AI tool and 88.2% of non-technical ads with no AI reference (“fewer than one in eight” mention AI). He recalls an earlier run at about 92% with no AI reference. The campaign headline he cites is “only 8% of UK businesses are hiring for AI skills”. The March 50-job test gave 2% explicitly requiring AI skills and 8% mentioning AI anywhere, and the April chat-vs-Cowork demo gave 7%. Treat any single run as one sample, and publish the methodology with the number.
Implementation
- Tool/Service: Claude Cowork (any paid plan; desktop app at the time of recording), Apify connector, optional Vibe Prospecting connector, Press Ranger for distribution.
- Setup: Apify account plus API token pasted into the Cowork connector; business-context files or memory loaded before ideation; a working folder for the spreadsheet and outputs.
- Cost: Apify gives 1 for a 1,000-job LinkedIn scrape. Vibe Prospecting: 400 free credits on sign-up (his figure). Press Ranger pricing is not given in the video. Claude usage comes out of the plan’s limits.
- Integration notes: the step prompts are in his free AI Recipe Vault (“AI press study” recipe). LinkedIn’s terms prohibit automated scraping; the Apify recipe article covers that caveat.
Try It
- Pick a question in your niche where the assumed answer may be wrong. For a dental marketer, for example: how many practice job ads mention any patient-communication software or AI tool? Write down the headline you expect before collecting anything.
- Run a 50-row pilot. Connect Apify, scrape 50 records into a spreadsheet, and check the columns and duplicates before paying for a large run.
- Scale to 1,000+ only once the pilot’s columns are right, keeping the raw spreadsheet as the reference copy.
- Analyse, then fact-check in a fresh prompt. Ask Cowork to recompute the headline numbers from the spreadsheet and show its method. Re-run the analysis once to see how stable the number is.
- Ship the data page before the press release, so every placement has something to link to.
Open Questions
- Placement quality. How many of the 381 placements are syndicated copies of one wire release rather than separate editorial coverage? The video doesn’t say.
- Which numbers were published? The video shows several figures (0.9%, 8%, 88.2%, 92%), and the transcript doesn’t make clear which one the coverage carried.
- Press Ranger cost and terms are not stated.
- Scale beyond one run. He mentions having analysed “nearly 2,000” roles across runs, while the run on screen kept 956. How the runs were combined is not shown.
Related
- Cowork + Apify Scraping Recipe — the same author’s step prompts for scraping and vibe prospecting, from his Notion recipe
- Claude Cowork Field Guide (Eliot Prince) — his general Cowork setup tips
- Cowork for Marketing — other marketing use cases for Cowork
- Claude Cowork (Product Overview) — the product and its current state
- AI Recipe Vault — where the prompts for this workflow are published
- Similarweb — Information Gain for AI Search — why original data earns citations