What I want from
the new site

Local dominance in search across their target market

One sentence: when a woman in the South Metro searches for help with a pregnancy, Pregnancy Choices is what she finds, and the site answers her question fast enough that she picks up the phone.

Three things follow from that, and everything in this document serves one of them.

• Local dominance. Apple Valley, Lakeville, Rosemount, Burnsville, Eagan, Bloomington, St. Paul, Minneapolis. Confirmed with the client 6 August. We rank per city, not on a national average.
• Fast and mobile-first. Nine in ten visitors are on a phone. Mobile is not a breakpoint, it is the product.
• Designed for the woman on the other end. Caring, not clinical. Helpful, not promotional. She may be frightened, she may be browsing at 2am, and she may not have long.

Metrics, with baselines, from day one

Build the dashboard before launch, not after. Every number below has a baseline already, which means we can show movement rather than assert it.

Lanes and the content chain

Claim the items in your lane, work them into tasks with Claude, and bring me specifics rather than status.

The content chain

Four stages. Most pages should never reach stage three.

Get her where she needs to be

Search is a primary feature, not a utility

Someone typing "how much does an abortion cost" is telling us exactly where she is. Most sites waste that moment. We are not going to.

Make it visible

A real input field in the header — not a magnifying glass icon. Placeholder does work: "Ask us anything — it's private."

Pre-seeded suggestion chips before she types. Tappable, and they double as a statement of what this place does:

Make it smart — intent-mapped calls to action

Results carry a contextual action, not just a list of links.

Make it forgiving

• Synonyms and misspellings — people type badly under stress. "plan b", "morning after", "pill to stop pregnancy" all resolve to the same place
• Spanish terms route to the /es/ equivalents
• Zero results never dead-ends — always falls through to call, text or schedule

Make it feed the programme

Fire the GA4 site_search event and review it monthly. This is free keyword research from precisely the audience we are trying to reach — better than any third-party tool, because it is actually them.

Every page ends somewhere useful

• Tap-to-call and tap-to-text on every page
• Schedule an appointment as its own persistent CTA, separate from Contact — they are different intentions
• Online scheduler that completes without a phone call, for the 2am browser
• Chat widget — LeadConnector exists today; decide whether it carries over
• Doxy.me video integration on the abortion pill page
• Map and directions with working event tracking (currently fired four times in two years)
• Separate thank-you URL per form
• Trust strip: free · confidential · no insurance needed · walk-ins welcome

Nine in ten visitors, so this is the product

• Bottom action bar, not top. Call · Text · Schedule in the thumb zone
• Pre-filled SMS body — sms:+19522349850?&body=Hi, I'd like to schedule an appointment. Removes the hardest part of texting a stranger
• Native map deep links rather than an embedded iframe — faster, opens the app she already uses
• No interstitials or pop-ups. The ays-popup-box plugin is installed today. Google penalises them and on this site they are actively hostile
• Correct input types — tel, email, inputmode, autocomplete. Fewer required fields; every one costs completions
• 44–48px tap targets, 16px minimum body text
• No carousels
• Test on a real low-end Android, not an iPhone simulator. That is what a lot of this audience is carrying
• Review the design in a dark room — traffic skews late and contrast behaves differently at night

Build once, reuse everywhere

Non-negotiables

Multilingual

• TranslatePress carried over and verified against Ollie before anything else locks
• Human-reviewed Spanish on the top ~15 pages — machine translation of medical content is a liability
• Translated slugs in the sitemap

The client wants movement. Here is how to do it right.

Ollie has animation support built in, so the capability is there. The discipline is what matters.

• Transform and opacity only. Anything animating height, width or position causes layout shift and destroys CLS — already our failing metric
• 150–300ms. Longer reads as slow, not smooth
• Never above the fold. Hero, phone number and primary CTA appear instantly. No exceptions
• Entrance only, once. Elements fade up as they enter view and do not re-animate on scroll-back
• No parallax on mobile. No scroll-jacking. Ever.
• prefers-reduced-motion respected — an accessibility requirement, not a preference

Formatting so machines can find and quote us

Some of this is real and some is hype. The real parts:

• Answer-first structure. Question as the H2, direct answer in the first sentence or two, detail after. This is what gets extracted, by Google's AI overviews and by ChatGPT
• Questions phrased the way she asks them, not the way a marketer would write them
• Real FAQ schema on every service page — not just an accordion that looks like one
• Named authors with credentials, a reviewer, and a review date. The strongest signal available on health content
• Clean server-rendered HTML. If content only appears after JavaScript runs, some crawlers never see it. Verify explicitly
• Consistent NAP in structured data

llms.txt — the Semrush audit flags it as missing. It is an emerging convention rather than an adopted standard, and no major AI company has committed to reading it. Cheap to add; do not expect much from it.

How much is too much, and yes we need an editor

This is the highest-risk content category there is: medical facts, legal exposure, and someone reading at 2am while frightened.

Google does not penalise AI content as such — it penalises unhelpful content. But health content gets the heaviest scrutiny, and AI drafts fail here in two specific ways: they get clinical details subtly wrong, and they produce emotionally generic prose. Generic is a failure mode on these pages.

Every clinical page gets named review with a date before it publishes. That is both an editorial standard and a ranking input.

Check every one lands

Heather owns this. Walk the signed proposal against this brief and confirm each deliverable is either built into the plan or consciously renegotiated with the client. Nothing gets quietly dropped, and nothing gets quietly added.

Particular attention to two:

• Content depth. The proposal's Phase 2 line is the only place content production is funded. At a 208-word median, bringing service pages to a competitive standard is a writing project, not a migration task. Size it honestly.
• Spanish. 103 pages, machine translated. Neither the proposal nor the research doc mentions it. Decide whether it is in scope and at what quality standard, and get that in writing.

Load the files, build your tasks, bring me specifics

All three of you: start a Claude conversation, load the supporting files, and work your lane into a task list. Do not start from a blank page — the analysis is already done.

Then

1. Work your lane in this brief into concrete tasks — with owners, estimates and dependencies
2. Generate an Avaza import file from the task list. Claude knows our import format: exactly 15 columns, Task Type between AssignedTo and Task Status, RowID and ParentRowID at the end. Company Name and Project Title must match the existing records character for character. Never open the CSV in Excel before uploading — it strips trailing columns and breaks the importer
3. Bring me the specifics: what you are building, what it costs, what you need decided