10. Getting the Digest Ready to Show Buyers

Summary

Closing out the trial and building the offer page

In the example, one invented customer is paying for a trial, which does not say where the second customer comes from. The chapter builds three things for that: an offer page, a short demo, and a record of who to contact next. The customer, runs, prospects and money are all still the show's invented example, and nothing here proves real demand.

Before building, the example's trial week is checked. The completed-runs record should hold five keys, one per scheduled date with no repeats. The customer record's delivered tally should be up by five, with the late-day credit field still empty. The two files are written by different steps, so agreement is evidence. They agree, the invented heating engineer says yes in writing, and the reply is logged with its date and how it arrived.

The page is drafted in Claude Code from the scope file alone, with one audience (single-owner trade businesses), one offer (the end-of-day digest) and one action (book a ten-minute look at a sample). The claim-check flags a line about winning back jobs as unsupported, so it is replaced with a plain description. No testimonial is used, since none exists.

Checks by hand cover contrast ratios under WCAG 2.2 (the WCAG 2.2 quick reference), heading order, image descriptions, and tabbing through the page. Lighthouse's accessibility score covers only machine-checkable audits, and WAVE cannot judge description quality, so neither replaces the manual checks.

Publishing with GitHub Pages

The page lives in its own small repository and is published from a branch. Pages sites are public, so the customer record, payments and notes stay out of that repository. Pushes made with a workflow's built-in token do not trigger a build. A custom domain is entered in Pages settings first, then DNS is set, and propagation can take up to a day.

A captioned demo

The demo is about a minute, narrated from a script and recorded in Clipchamp using the labeled sample day, with "invented sample data" on screen throughout. Automatic captions need a voice track and misfire on key words, such as "digest" becoming "diet chest", so each line is corrected by hand. Prerecorded captions are a basic accessibility requirement. Whether to burn captions in or ship a separate file is left to check against current guidance.

Prospects in HubSpot

HubSpot's free CRM holds three invented prospects (a plumber, a roofer, an electrician), each with a company and a dated follow-up task, through the HubSpot connector for Claude. It needs a paid Claude plan and admin approval in HubSpot. HubSpot also offers a separate remote MCP server. Write permissions and Claude Code setup should be checked in current documentation.

The agent's report is not trusted: each record is opened in HubSpot. In the example, one task had no due date and is fixed and rechecked. None of the three pieces has met a real owner yet.


In the example, one invented customer is paying for a trial. That does not tell you where the second customer comes from. This chapter answers that with three things you can build in an afternoon with Claude Code and a couple of ordinary tools. The first is one honest offer page with one clear next step. The second is one short demonstration a stranger can watch. The third is one record of who to contact next and what you promised them. Everything about the business is still the show's invented example: the customer, the runs, the prospects and the money. Nothing here proves real demand. That only comes when a real owner sees the page and commits.

The offer page, published

Close out the trial before you build anything. When we left the example, the two-hundred-and-seventy-dollar monthly offer was drafted and held back until the cloud scheduled task ran a clean week with someone watching it. So here is that check, done the way the earlier lesson set it up. Say the five invented weekday runs all show in History. Open the completed-runs record and confirm there are five keys, one for each scheduled date, with no date repeated. Then open the customer record. Its delivered tally should have gone up by five, and the late-day credit field should still be empty. The two files are written by different steps, so if they agree, you have evidence the week happened as the history claims. In the example they agree. The offer goes out with its terms intact: monthly billing starts only once the customer agrees in writing. The invented heating engineer replies yes. You log that reply, with its date and how it arrived, in the customer record before anything else happens. That written yes is what the offer is waiting for, and now the business has something worth showing to the next person.

Now the page. Open the Digest Prototype folder in Claude Code and switch to plan mode. Ask for a single file called index, an ordinary web page. Name the evidence-checked-drafting skill, and say the page must be written only from the scope file. That file already says who the digest is for, what arrives by seven each weekday evening, the thirty-enquiry ceiling and everything it will not do. Tell it the page has one audience, small trade businesses with one owner who miss enquiries while on site. It has one offer, the end-of-day digest at its stated price. And it has one action: book a ten-minute look at a sample digest. These three rules come from people who build landing pages for a living. They are rules of thumb, not a standard. They suit this stage because a stranger should finish the page knowing exactly what to do.

Read the plan, approve it, and read the draft's claim-check list before you read the page. In the example, the draft has a friendly line under the headline that says owners "win back the jobs they used to lose." The claim-check marks it unsupported. No source file says any job was won back. The trial customer received digests, and that is all the record shows. Delete the line. Its replacement is plainer and true: every enquiry from the day, on one page, with the unreachable ones first. Proof works the same way. You have one invented customer and no real testimonial, so the page carries none. A labeled sample digest is something you can honestly show.

Then check the page by hand, because the page has to work for the people reading it. Start with contrast, meaning how far the text color stands apart from the background. The accessibility guidelines, version two point two, ask for at least four and a half to one for normal text and three to one for large text. Ask Claude Code to calculate the ratio for every text and background pair in the stylesheet, then confirm the lowest one yourself in a contrast checker. Next, read the heading outline in order: one main heading, then sections beneath it that make sense read alone. Then judge each image description. If the sample digest appears as a picture, its description should say what a listener would need, for example that it is an invented sample showing seven enquiries with one marked unreachable. A description that just says "screenshot" fails. Finally, put the mouse down and press Tab through the page. Every link and the booking button should get a visible outline, in a sensible order.

Free tools help, but they are not the whole check. Chrome's Lighthouse gives an accessibility score that covers only the audits a machine can pass or fail, so a high score does not mean the page is accessible. WebAIM's WAVE flags errors and alerts, but it cannot tell whether an image description is any good, and its own help says that no errors does not mean the page is accessible. Run them, fix what they find, and keep the four checks you did by eye.

To publish, use GitHub Pages, which hosts a static page straight from a GitHub repository. Choose it over a website builder because the page is already a file you keep in git. Give it a separate small repository that holds only the page and its images, for a reason explained in a moment, then commit index to the top of that repository and push. In the repository on GitHub, open Settings, then Pages. Under Build and deployment, set Source to Deploy from a branch, choose your main branch and the root folder, and save. Changes can take up to ten minutes to appear. Then click Visit site, or open the address, and read the live page. Don't stop at the copy on your laptop. If nothing shows after an hour, the usual cause is a build error. GitHub runs a site generator called Jekyll by default, so look in the repository's Actions tab for a failed run. One trap matters for agents. If a GitHub workflow pushes the page using its built-in token, that push does not start a Pages build. So have Claude Code push from your own session, or check that a build ran.

The reason for that separate repository is one fact about Pages. A Pages site is public, even when the repository is private. On a free account the repository has to be public too. Keeping a site private needs an enterprise plan, and the plan rules shift, so check GitHub's current plan table if that matters to you. In practice, that means the customer record, the payments record, notes files and anything with a real name stays out of whatever folder Pages publishes. Keeping the page in its own repository is the simplest guard.

A proper address comes later. If you want your own domain, enter it in the Pages settings first, then set the DNS records at your registrar. Use A records for a bare domain, or a CNAME pointing a subdomain at your github.io address. DNS changes can take up to a day to spread. HTTPS comes through a free certificate, and the Enforce HTTPS box can also take up to a day to appear.

A captioned demonstration and a prospect record

The page tells a stranger what the digest is. The demonstration lets them watch one evening's result arrive. Make it from the labeled sample, the invented day that passed all five criteria. A stranger can then see the real output format without anyone's actual enquiries on screen. Write the narration first, as a short script of about a minute. Say the problem in one sentence, open the folder of the day's invented files, show the digest page with the unreachable enquiry at the top, point to the count line matching the file count, and finish on the booking action. Put "invented sample data" on screen from the first frame.

Record it with Clipchamp, Microsoft's video editor, which is built into Windows and also runs in a browser. It has a screen recorder and automatic captions in one place, so there are no files to pass between tools. Record the screen while you read the script aloud. Narration does more than explain. Clipchamp's automatic captions only work on video or audio that contains a voice, so a silent screen recording gets none. Turn on auto-captions, then read the whole caption track against your script line by line. This matters for more than polish. The accessibility guidelines require captions for prerecorded audio at their most basic level, and captions that drop dialogue count as a failure. Automatic captions most often go wrong on exactly the words you care about. In the example, "digest" comes out as "diet chest" and "Drive" as "drives". Fix each one by hand. Export as an MP4 file at ten-eighty p, the resolution Microsoft recommends, and keep captions switched on in the export, because captions you turned off do not come out. You now have a finished, captioned demo. One choice is still yours to check against current guidance: whether to burn the captions into the picture or ship a separate caption file, and whether to put the video directly in the page or link to a hosted player.

Now the record of who to contact. A customer relationship management system, usually just called a CRM, is the business's list of people, the deals in progress with them and the next action owed to each. A spreadsheet can do this for three names. It breaks when follow-ups pile up and an agent needs to update it reliably. HubSpot's free CRM is a good start here. HubSpot makes a connector for claude.ai, and it works on every HubSpot tier, the free one included, though it needs a paid Claude plan. Features from HubSpot's paid hubs, such as campaigns and quotes, are not available on free. A HubSpot account with admin rights has to approve the connection first. After that, the agent sees only what your own HubSpot permissions allow, and properties HubSpot marks as sensitive are excluded.

Turn the connector on in claude.ai, then give it one specific request. Create three illustrative prospects, labeled invented in a note on each record. Say they are a plumber, a roofer and an electrician, each a contact with a company. For each one, add a task due on a stated date, such as sending the page and the demo, then checking for a reply after three working days. HubSpot's connector page lists creating and updating contacts, companies and deals, and logging notes and tasks. The exact permissions behind those write actions, and how to set up the same connection inside Claude Code, are worth checking in the current Claude and HubSpot documentation before you depend on them. HubSpot also runs a separate remote MCP server. It is a different route from the claude.ai connector, and Claude Code would reach it through MCP.

Then check the destination, not the reply. The agent will say it created three prospects with tasks. Ask it for each record's identifier, then open HubSpot yourself and search for each one. Confirm the contact, the company link, the note saying invented, and a task with the right due date on the activity timeline. In the example, two match. The third task landed on the contact but with no due date, which would have quietly hidden the follow-up. Add the date, ask the agent to recheck that record, and look again.

Now try it once on your own offer, away from the example's files. Write down who it is for, what they get and the one thing you want them to do next. Ask Claude to draft a single page from only those notes and to list every claim with the line that backs it. Cut each claim with nothing behind it. Then press Tab through the draft and check that every link lights up in order. You've done it well when every sentence left points to something you can show, and the page ends on that one action.

That leaves the business with a public page that makes only claims its scope supports, a captioned demo built from labeled sample data, and a CRM that holds every prospect's next action where you can inspect it. All three were checked where they actually live. None of them has met a real owner yet. The next real step is putting that page in front of one, and the CRM is ready to keep track of what you promise when you do.

Want a Series of your own?

Gnothi wrote and narrated this chapter. Name a subject and it writes you a course, or a story that teaches it, one chapter at a time, delivered to your podcast app.

Get started free