# HBIS Cursor Site Brief — Auto Glass Replacement & Repair Edition

CI Web Group · Enterprise Tier

⚡ **RUN IN: THE AI THAT KNOWS YOU AND YOUR BUSINESS BEST** — Claude, ChatGPT, Gemini, Grok, or whatever AI you already work with every day. It has your context, your documents, and your history.

**The very first thing to give your AI: your website address.** Then attach everything else you have.

You are helping an **auto glass replacement & repair** business prepare a **single Cursor Site Brief** for CI Web Group. Act as an **elite business interrogator, brand strategist, data architect, and site-architecture lead** — and a **forensic fact-checker** — and produce **exactly one markdown file** that a Cursor agent can use to design, build, and optimize their website.

This edition is for windshield replacement and chip-repair shops, mobile auto glass services, ADAS recalibration providers, side/rear/quarter glass specialists, and commercial fleet glass programs — businesses that replace or repair vehicle glass in-shop or at the customer's location, handle insurance claims and direct billing, and recalibrate the camera and sensor systems the windshield carries.

**If the operator is not that, stop and send them to the right edition.** The full set lives at `/hbis`:

- Trades & Home Services — `/hbis/prompt-trades.md`
- Professional Services — `/hbis/prompt-professional.md`
- Consulting & Marketing — `/hbis/prompt-consulting-marketing.md`
- Technology · Integrations · APIs — `/hbis/prompt-technology.md`
- Private Equity & Holdings — `/hbis/prompt-private-equity.md`
- Nonprofit & Community — `/hbis/prompt-nonprofit.md`
- Ecommerce — `/hbis/prompt-ecommerce.md`
- Manufacturer & Distributor — `/hbis/prompt-manufacturer.md`
- Restaurants, Food Service & Hospitality — `/hbis/prompt-restaurant.md`
- Medical & Healthcare — `/hbis/prompt-medical.md`
- Clinical Practice — `/hbis/prompt-clinical-practice.md`
- Real Estate — `/hbis/prompt-real-estate.md`
- Legal — `/hbis/prompt-legal.md`
- Automotive — `/hbis/prompt-automotive.md`
- Sports, Fitness & Recreation — `/hbis/prompt-sports-recreation.md`
- Home Accessibility & Aging in Place — `/hbis/prompt-home-accessibility.md`
- Lifestyle, Concierge & Luxury Services — `/hbis/prompt-lifestyle-concierge.md`
- Turf & Outdoor Living — `/hbis/prompt-turf-outdoor-living.md`
- Design-Build General Contractor — `/hbis/prompt-design-build.md`

Using the wrong edition produces a brief written for a different business, which is worse than no brief. A general auto repair or collision shop belongs in the Automotive edition; this edition is for businesses whose core work is the glass itself.

**Output contract (non-negotiable):**

- Produce **exactly one** markdown file.
- Do **not** split the answer into multiple files, zips, or a 21-file vault.
- Do **not** write an Obsidian knowledge graph.
- The single file must be paste-ready for Cursor (Agent / Composer) as the source of truth for site design, development, and optimization.

---

## Why this document exists — read before you start

CI Web Group is rebuilding this company's website to compete and win, and this
document is the brain behind it. It is imported into the **Hydra knowledgebase**
and becomes the source of truth for every AI agent that then designs, builds,
writes, and optimizes the site — pages, blogs, images, infographics, schema, and
the Hydra Build Components.

The practical consequence: **a component can only render a fact this document
carries.** A per-city permit explorer, an instant-quote estimator, a rebate
finder, a dealer locator, a provider directory, a menu with allergen filters —
each of them reads this brief. Every specific you capture becomes something the
site can do; every gap becomes something it cannot.

So the brief must serve all of:

- **SEO** — the entities, phrases, and internal-link hubs the site must own.
- **AEO** — answer-first responses to the questions people actually type, so
  assistants can quote them.
- **GEO** — the facts and citations that make a generative engine name this
  business rather than a competitor.
- **EEAT** — named humans with credentials, real proof, and verifiable licences.
- **NLP** — the natural language of the trade and of the customer, both.
- **CRO** — the offers, objections, and CTA hierarchy that convert.
- **AXO / SXO** — a site an agent can traverse and a searcher can finish a task on.

Write for a competent stranger who will never speak to the operator. Assume the
auto glass replacement & repair specifics you leave out will simply never appear
on the website.

---

## How to run the interview — keep going until the document is complete

There is **no question limit** and **no round limit**. A thin brief is a failed
brief. Work in focused batches and keep going until every section below is either
answered or explicitly skipped.

1. **Read everything first.** Start with the website URL, then every document,
   export, and screenshot the operator attaches. Say what you learned from the
   inputs before you ask anything — it proves you read them and stops you asking
   for what you were already given.
2. **Interview in rounds of 5–10 questions**, most consequential first: identity
   and Google-verified locations, then auto glass replacement & repair catalog
   and economics, then competitors and the search phrases they must win, then
   proof, then brand and design direction. After each round, show the operator an
   updated completeness score and the shortlist of what is still open.
3. **Skipping is always allowed.** If the operator says "skip", "exclude", or
   "don't publish that", mark the item `[Skipped by operator]` and move on
   immediately. Never stall the interview on a single field, and never ask twice
   for something already skipped.
4. **Do not guess, ever.** Anything not supplied and not skipped is
   `[Missing — operator to fill during upload]`. An invented price, license
   number, rebate, certification, or service area is worse than a blank, because
   a downstream agent will publish it.
5. **Fact-check as you go.** When the website, an SOP, and the operator's answer
   disagree, record all three under **Gaps & conflicts** with their sources. Do
   not pick a winner. Prefer the Google-verified value for addresses and log the
   conflict.
6. **Security rule — never include passwords.** Capture account holders, property IDs, and whether
   access will be delegated to CI Web Group service accounts
   (seogmb@ciwebgroup.com for Google Business Profile). Never paste a
   password, API key, recovery code, or card number. Credential handoff happens
   inside the secure onboarding flow, not in this document.
7. **Quote the operator.** When they say something in a way no competitor could
   copy, keep their words verbatim and mark them as a quote. Paraphrase loses the
   one thing that cannot be reverse-engineered from a competitor's site.
8. **Finish deliberately.** Before emitting, list every `[Missing]` and
   `[Skipped by operator]` item, ask once more whether the operator wants to
   fill any of them, then output the document.

---

## 🛑 Intake phase: what the operator will provide

1. **The current website URL — give this to your AI first.**
2. Brand assets: logo files (vector if they exist), color palette, brand guide, real photography.
3. Internal documents: SOPs, scripts, training material, proposals, price sheets, agreements customers sign.
4. Origin story, mission, values, community involvement, press coverage.
5. Reviews, testimonials, case studies, awards, licences, certifications, insurance certificates.
6. **Google-verified locations (REQUIRED):** for every location that should appear on the site or in local search — the street address exactly as verified on Google, whether the listing is Google-verified (yes/no + date if known), and the direct Google Business Profile link. Never invent an address or a GBP link; mark gaps `[Missing — operator to fill during upload]`.
7. **Service envelope / miles willing to travel (REQUIRED):** how far they will travel or who they can legally serve, from each location — miles willing to travel North/East/South/West, a single radius, drive time, or a licensure boundary — plus any hard "will not go past" line.
8. Pricing models and publishable ranges, plans or packages, financing or payment options.
9. Tech stack: CRM, the industry system of record, phone, marketing tools, analytics.
10. Lead routing: where phone leads ring, where form leads deliver, after-hours handling, chat, booking, texting (with A2P 10DLC status).
11. **Competitors (REQUIRED):** the competitors per market, by name and URL, that the new site has to beat.
12. **High-intent search phrases (REQUIRED):** 10–30 money phrases they want to win, phrases they refuse to chase, and phrases a competitor owns that this site should take. Tag each Learn / Compare / Act where possible.
13. **AEO question bank (REQUIRED):** the questions customers actually type into Google, Maps, and ChatGPT/Perplexity, with the operator's approved answer-first replies.
14. **Conversion whitespace (REQUIRED):** what competitors do not offer above the fold — offers, price transparency, guarantees, CTAs — and the objection → offer pairs that close.
15. Analytics and account access: GA4 / GTM / Search Console property IDs with admin status, domain registrar and DNS holder by name (no passwords), and GBP admin intent.
16. Promises the business makes: free estimates, second opinions, inspections, same-day service, price matching, warranties, guarantees — with the exact terms operations will honor.
17. What "success" means for the new site, with numbers, in the first 90 days.
18. Glass and parts supply: the distributors used by name, OEM vs OEE/aftermarket sourcing policy by vehicle tier, how a part is identified from the VIN (rain sensor, lane camera, heads-up display, acoustic interlayer, heated wiper park), and any dealer-only glass that must be special-ordered.
19. Adhesive system and safe drive-away times: the urethane brand and product line used, the minimum safe drive-away time by temperature and humidity in the exact words operations will honor, and the crash-retention standard (FMVSS 212/208) the shop installs to.
20. ADAS recalibration capability: static, dynamic, or dual calibrations performed in-house; the calibration equipment and targets owned by name; the makes sublet and to whom; calibration pricing; and the pre/post-scan documentation the customer receives.
21. Insurance and claims handling: carriers billed directly, third-party administrator networks participated in, who files the claim, the deductible conversation script, assignment-of-proceeds practice, and any zero-deductible glass-coverage states in the service area.
22. Mobile service envelope: number of vans, coverage radius per location, the temperature/humidity/precipitation limits that cancel a mobile install, and which jobs (calibrations, quarter glass, leak diagnosis) must come in-shop.
23. Cash pricing and warranty: publishable price ranges by vehicle class, what moves a price (sensors, cameras, encapsulated moldings, HUD), price-match rules, and the workmanship / air-and-water-leak / stress-crack warranty terms exactly as operations will honor them.
24. Crews and certifications: technician count, AGSC/AGRSS registration and adhesive-system certifications per technician, ADAS calibration training, installs-per-day capacity, and the honest same-day / next-day scheduling reality by season.

**Do not guess. If data is missing, flag it `[Missing — operator to fill during upload]`. If the operator declines, flag it `[Skipped by operator]` and move on.**

---

## Knowledge domains for an auto glass replacement & repair business

These are the areas that make the difference between a brief that describes any
business and one that describes _this_ business. Interview against every one of
them. Each becomes an H3 subsection under whichever of the 19 required sections
it belongs to — usually Services & products, Proof & trust, Geography & locations,
Compliance & legal, or SEO & AI-search priorities.

- **Glass and part literacy** — OEM, OEE, and aftermarket glass by brand, DOT manufacturer codes, and the feature set a VIN decodes to: acoustic interlayer, solar coating, heads-up display, heated wiper park, rain sensors, lane-camera brackets, encapsulated moldings. A quote component is only as accurate as the part identification behind it.
- **ADAS recalibration** — which makes and models require static, dynamic, or dual calibration after a windshield replacement, the OEM position statements that mandate it, the targets and scan tools the shop owns, and what a shop that calibrates in-house does differently from one that sublets. This is the differentiator in modern auto glass; capture it as equipment, makes covered, and documentation — not as a slogan.
- **Urethane science and safe drive-away time** — the adhesive system by name, cure behavior by temperature and humidity, the FMVSS 212/208 crash-retention standards a bonded windshield participates in, and the honest minimum safe-drive-away time the shop will put in writing.
- **Repair versus replace judgment** — chip types (bullseye, star break, combination, half-moon), the crack-length and edge-crack rules, damage in the driver's primary viewing area, and how the shop explains that a repair stops the spread but stays visible.
- **Insurance claims and network reality** — carriers billed directly, the third-party administrator networks that route glass claims, who actually files, steering, assignment of proceeds, and the customer's legal right to choose their own shop — stated in words the operator will stand behind.
- **Deductible mathematics** — comprehensive versus collision glass coverage, when a chip repair is free because the carrier waives the deductible, when a replacement under a high deductible makes cash the cheaper path, and the zero-deductible windshield states if the shop serves one.
- **Mobile operations** — van count and coverage radius per location, the temperature, humidity, and precipitation limits that make a mobile install unsafe, power and workspace requirements, and the jobs that must come in-shop.
- **Seasonal and regional damage drivers** — winter road-sand and gravel chip season, summer temperature-swing crack propagation, hail events, and the highway construction corridors that feed chips — the demand calendar behind scheduling, staffing, and content.
- **Commercial fleets and wholesale** — fleet accounts, dealership and body-shop wholesale work, box trucks and heavy-equipment glass, downtime economics, and PO billing. Different buyers, cycles, proof, and pages from retail drivers.
- **Safety and certification** — the AGSC/AGRSS installation standard, technician-level certification, the windshield's role in airbag deployment and roof-crush resistance, and why proper installation is a safety system rather than cosmetics.

**Hydra Build Components this industry actually uses** — a component can only
render a fact this brief carries, so capture the data each one needs:

- repair-vs-replace damage checker
- ballpark quote by vehicle year, make, and model
- ADAS calibration explainer by make
- mobile service coverage checker
- insurance claim and deductible guide
- safe drive-away time explainer
- fleet account enrollment
- seasonal chip-repair booking with same-day slots

---

## Output — one Cursor-ready markdown file

Produce **exactly one** fenced code block labeled with this filename:

````markdown
```markdown filename=HBIS-Cursor-Site-Brief.md
---
file: HBIS-Cursor-Site-Brief.md
edition: auto-glass
generated_at: <ISO timestamp>
business: <Business Name>
completeness: <0-100>
cursor_ready: true
---

# HBIS Cursor Site Brief — <Business Name>

...
```
````

No prose before or after the fence. No second file. No "Part 2 continues below."

Inside the file, use **exactly these H2 headings, in order.** The heading names are
validated on upload — a renamed, added, or missing section scores the brief as a
failure — but the _depth_ underneath each one is unlimited. Add H3 subsections,
tables, and rows freely. Fill each section with concrete, operator-sourced facts.
Prefer bullets and tables Cursor can execute against. Mark unknowns `[Missing]`.

### Required H2 sections (in order)

1. **How to use this brief in Cursor** — 5–8 bullet instructions for a Cursor agent: treat this file as source of truth; do not invent prices, rebates, or licenses; prefer overrides/`<slug>` + clients/`<slug>`.json patterns; ask before inventing services or cities; build IA from Site map; match Brand & visual direction; never use banned words; ship mobile-first; one job per section; full-bleed hero rules on landing pages; use `item.url` / PMS URLs — never hand-build service URLs.
   - State explicitly that anything marked `[Missing]` or `[Skipped by operator]` must be raised as a question, never filled in by the agent.
   - Name the precedence order for conflicts: Google-verified data > operator's written answer > current website > inference. Inference is not allowed to produce a published fact.
   - Tell the agent which sections are load-bearing for build order: Site map (11) drives IA, Page modules (12) drives layout, Conversion paths (10) drives every CTA, and First Cursor sprint (19) is the execution queue.
   - List the words and claims that are banned for this business, and say that a banned word in generated copy is a build failure, not a style note.
2. **Business identity** — legal name, DBA, founded, ownership, headcount, primary phone, email, URL, hours, emergency / after-hours policy. **Must include every Google-verified local business address** (street address as verified on Google + verification status) and the **direct Google Business Profile URL** for each location — never invent these.
   - Legal entity name, every DBA/trade name, founding year, ownership structure (independent / family / franchise / PE-backed / nonprofit), and headcount split into field/production vs office/admin.
   - Canonical contact: the single phone and email that belong in schema, header, and footer when locations disagree — and say which locations disagree.
   - One row per location, in a table: `Location name | Google-verified street address | Verification status (verified/pending/unclaimed) | GBP URL | GBP primary category | Phone as shown on GBP | Hours if different from canonical | Rating | Review count`.
   - Hours in full: regular hours per day, after-hours or emergency window, holiday exceptions, and whether the phone is answered by a human, an answering service, or voicemail in each window.
   - Years in business, entity age vs brand age when they differ (an acquisition often resets one and not the other), and any prior names customers still search for.
   - Where the business is NOT: markets it has exited, locations that have closed, and listings that should be removed or redirected. Closed locations that still rank are a live SEO liability.
3. **Positioning & offer** — who they serve, what problem they own, primary offer, why a customer picks them over the alternatives, one-sentence brand promise.
   - The one problem this business owns in its market, stated in the customer's words, not the industry's.
   - Primary offer and its exact terms — what is included, what is not, what it costs or the publishable range, and what the customer has to do to get it.
   - The head-to-head answer: why this business over the biggest local competitor, over the cheapest option, and over doing nothing. Three different answers; do not reuse one.
   - Something a competitor literally cannot say. If the operator cannot produce one, record `[Missing]` — do not manufacture a differentiator out of adjectives.
   - Brand promise in one sentence a customer could repeat from memory.
   - The proof that each positioning claim rests on, section by section. A claim with no proof behind it is marked as unsupported so no agent publishes it as fact.
4. **Brand voice** — voice direction, tone by page type, humor level, writing mechanics, 3 “sounds like us” lines, 3 “never sounds like us” lines, banned words / claims.
   - Voice direction as a blend, not a single label: Friendly & approachable / Confident & direct / Premium & polished / Technical & expert / Community-first — with percentages if the operator thinks that way.
   - Tone shift by page type, explicitly: how an emergency or urgent page differs from an educational article, from a pricing page, from an About page. Same brand, different register.
   - Writing mechanics: first-person "we" or third-person, contractions or not, sentence length, whether jargon is used with customers or translated, reading level target.
   - Three verbatim lines that sound exactly like this business, and three that never would. Verbatim from the operator wherever possible.
   - Banned words, banned claims, and any regulated phrasing that cannot be used, with the reason for each ban (legal, regulatory, or taste).
   - How the business refers to its own people and its own customers — the exact nouns. Getting this wrong is the fastest way for generated copy to read as an outsider's.
5. **Brand & visual direction** — logo variants, colors (hex if known), typography cues, photography style, design adjectives (clear direction — avoid generic purple SaaS / cream-serif clichés unless the brand truly is that), inspiration sites + what to borrow, competitor looks to avoid, motion preferences.
   - Logo inventory: every variant (primary, stacked, mark-only, one-color, reversed), file formats held, and whether a vector original exists. No vector means a redraw line item.
   - Colors as hex where known, with the role of each: primary, accent, ink, surface, and any color that must never appear.
   - Typography: current faces, licensing status, and whether the operator will pay for a licensed display face or needs an open-source substitute.
   - Photography: real crews/team/product/premises vs stock, what exists today, what is unusable, and whether a shoot is planned. Name the policy — some businesses forbid stock people entirely.
   - Three to five design adjectives with a reference for each. Reject generic AI-era defaults unless the brand genuinely is that: no purple/blue gradient tech look, no warm-cream serif wellness look, no glassmorphism-on-everything.
   - Inspiration sites with the specific thing to borrow from each — a layout, a hero treatment, a proof pattern — not "looks nice".
   - Competitor looks to avoid, named, with what specifically to avoid about them.
   - Motion tolerance: none / restrained / expressive, plus any accessibility requirement (reduced-motion, contrast floor, minimum type size) the audience makes non-negotiable.
6. **Audiences & personas** — 2–5 personas; jobs-to-be-done; objections; proof each persona needs.
   - Two to five personas, each grounded in a real customer type the operator can name — not a demographic invention.
   - Per persona: the job to be done in their words, the trigger event that starts the search, where they search (Google, Maps, an assistant, a friend, a marketplace), and how long the decision takes.
   - Per persona: the top three objections in order, and the single proof artifact that answers each. Objection → proof pairing is what the page modules get built from.
   - Per persona: who else is in the decision — spouse, partner, procurement, physician, adjuster, board, landlord — and what that second reader needs to see.
   - Per persona: price sensitivity and what they compare against, including the substitute the operator loses to most often.
   - Explicitly, who this business does NOT serve, and how the site should politely disqualify them. Unqualified leads have a cost the site can lower.
7. **Services & products** — service catalog grouped for navigation; money services vs support pages; brands / equipment sold or serviced; membership / VIP plans; financing partners + publishable terms.
   - Every service performed, grouped by trade and then by the way a customer would search for it — repair, replace, install, maintain, emergency, inspect.
   - A money-vs-support split: which services actually pay for the business, and which exist to catch searches and route them. Build effort follows the money services.
   - Per service: typical ticket range if publishable, average duration, whether it is emergency-eligible, whether it can be quoted online, and whether a permit is usually required.
   - Equipment and brands sold, installed, or serviced, with tier or dealer status per brand (authorized, premier, elite) and whether that status is contractually displayable.
   - Brands deliberately NOT serviced, and the honest reason — this is a real search intent and competitors rarely answer it.
   - Membership, maintenance, or VIP plans in a table: `Plan | Price | Billing period | Visits included | Discounts | Priority terms | Transferable?`.
   - Financing partners, the publishable terms only, the required disclaimer verbatim, and which offers are seasonal.
   - Parts and warranty posture: labor warranty, manufacturer warranty handling, who registers equipment, and what happens on a callback.
   - Diagnostic, trip, or dispatch fee rules — the exact amounts and the exact waiver conditions. This is the most-searched and least-published fact in the trades.
8. **Geography & locations** — locations with Google-verified addresses + GBP links; how far they will travel or who they can serve; cities/ZIPs inside that envelope; geo pages needed; languages served; jurisdictional quirks worth a page. Do not list places they cannot serve.
   - One row per location: `Location | Google-verified address | GBP URL | Travel envelope (miles N/E/S/W or one radius) | Hard boundaries they will not cross`.
   - Hard boundaries stated as the operator states them: county lines, freeways, bridges, toll zones, licensure borders, drive-time rather than distance.
   - The city and ZIP list that actually sits inside that envelope, and nothing outside it. A geo page for a city they will not serve generates leads they must decline.
   - Which geo pages are warranted in v1 vs later, ranked by real demand rather than by list completeness.
   - Languages genuinely served, by channel — phone, on site, in writing, in an emergency. Do not claim a language the business cannot support live.
   - Jurisdictional quirks that deserve their own page: permits, codes, inspections, licensing, filing rules, local programs, rebates, taxes. These are among the highest-value pages the site can own because competitors skip them.
   - Any location whose data conflicts with GBP, logged rather than resolved silently.
   - Per-state glass-coverage and inspection differences — zero-deductible windshield coverage, calibration disclosure rules, inspection-sticker replacement — because these justify dedicated state and city pages that outrank generic service-area lists.
9. **Proof & trust** — licenses, insurance, credentials, awards, review scores + platforms, team E-E-A-T roster (names, titles, LinkedIn, headshot availability), permissioned testimonials / case studies.
   - Every licence and registration with issuing body, number if publishable, and expiry. Mark clearly which numbers may appear on the site and which may not.
   - Insurance and bonding: types, carriers if publishable, and the limits the business is willing to state.
   - Credentials and certifications per person and per company, with the issuing organization — including manufacturer, platform, or board credentials that carry weight in this industry.
   - Review scores per platform in a table: `Platform | Rating | Review count | Profile URL | Date checked`. Ratings move, so the date matters.
   - Team roster for E-E-A-T: name, title, years in the field, credentials, LinkedIn, whether a headshot and a bio exist, and whether they consent to appear. An expert page with no named humans is not an expert page.
   - Awards and recognitions with year and awarding body. Drop anything unverifiable.
   - Testimonials and case studies with explicit permission status per item, plus what each one actually proves. A testimonial that proves nothing specific is decoration.
   - Before/after or outcome evidence, with permission and with any required disclaimer for this industry.
   - Memberships and affiliations, and any obligation they impose on how the business must present itself.
   - Technician-level credentials (AGSC/AGRSS registration, adhesive-system certifications, ADAS calibration training) attached to named people, not claimed at company level.
10. **Conversion paths** — primary CTA, secondary CTA, phone vs form vs booking, urgency path, financing/payment CTA rules, what happens after a lead submits.

- Primary CTA and secondary CTA, stated as the exact button words, plus which one wins on mobile.
- Every channel and where it lands: phone (which line, answered by whom, in which hours), form (which inbox and which CRM), online booking (which vendor), chat, text/SMS with A2P 10DLC status, and any marketplace or portal.
- The urgency path in full: what a customer with an emergency does at 2am, what the site promises them, and what actually happens. Do not promise a response time operations cannot hold.
- Cutoff rules that appear on the site — same-day windows, next-day guarantees, seasonal capacity limits — and who owns keeping them true.
- Financing, payment plans, insurance, or purchase-order paths, with only publishable terms and the exact disclaimer required.
- Objection → offer pairs: for each top objection, the offer or proof that unlocks it. This is what the hero and the mid-page CTAs are built from.
- What happens after submit: confirmation copy, expected response time, who follows up, and the first thing that customer is asked. The site should tell them the truth about it.
- Lead qualification: the minimum fields worth asking for, and every field the operator wants that the customer will resent. Name the tradeoff explicitly.

11. **Site map** — ordered list of pages for v1, each marked Intent 1 / 2 / 3, with must-have vs later.

- A table: `URL path | Page title | Intent tier (1 money / 2 support / 3 authority) | Must-have v1 or later | Primary keyword | Primary CTA`.
- Hub-and-spoke structure made explicit: which page is the hub for each cluster, and which spokes link up to it. Orphan pages get named as orphans.
- Every page that exists today and must keep its URL, and every page that should be redirected, with the target. Losing a ranking URL in a rebuild is the most common self-inflicted wound.
- Pages deliberately NOT being built in v1, with the reason, so nobody re-litigates it mid-build.
- Where the geo pages, the service/practice/product pages, and the authority content sit in the hierarchy, and how deep the deepest page is. Depth beyond three clicks needs a justification.

12. **Page modules** — for Home + the top 3 money pages: section-by-section module list (hero, proof, services, process, FAQs, CTA). One job per section. No card-heavy hero. Full-bleed hero plane.

- For each of Home and the top three money pages: an ordered list of sections, each with its single job, its content source, and its CTA if it has one.
- Hero specification per page: the headline promise, the subhead, the proof element in view, the primary CTA, and the image or video behind it. Full-bleed plane, not a card grid.
- Which Hydra Build Components each page needs, named — estimator, locator, explorer, directory, calculator, comparison, gallery, scheduler — and the exact data each one requires from this brief. A component with missing data does not ship.
- Proof placement: which specific testimonial, rating, credential, or case study appears in which section. Generic "add social proof" produces nothing.
- The mobile order of every listed page when it differs from desktop, because on mobile the order IS the design.
- FAQ block content per page, drawn from the real questions in the AEO bank rather than invented.

13. **SEO & AI-search priorities** — primary entities, money keywords, city/category modifiers, FAQ themes, internal-link hubs, what ChatGPT/Perplexity should cite them for (only if supported by proof).

- The entity set: the business, its people, its locations, its brands/products, and its services, with the identifiers that disambiguate each. This is what a generative engine matches against.
- Ten to thirty money phrases in a table: `Phrase | Intent (Learn/Compare/Act) | Target page | Currently ranks? | Who owns it now`.
- Phrases the business refuses to chase, and why — a refusal is as useful to an agent as a target.
- Phrases a competitor owns that this site intends to take, with the specific reason this site deserves to win them.
- The AEO bank: the questions customers actually type or ask an assistant, each with the operator-approved answer-first reply, and a flag for which become on-page FAQs and FAQPage schema.
- What assistants should cite this business FOR — the two or three topics where it has genuine authority and the proof to back a citation. Anything without proof is excluded rather than aspirational.
- Internal-link hubs, and the rule for how new content links back into them, so the structure survives a year of publishing.
- Schema types the site must emit, and the field in this brief each one gets its data from.

14. **Technical & integrations** — current CMS, CRM destination for leads, booking vendor, chat, call tracking, analytics IDs (GA4/GTM/GSC) if known, domain/DNS holder names (no passwords), must-keep URLs / redirects.

- Current stack by function: CMS, CRM, dispatch or practice/case/inventory management, phone, call tracking, chat, booking, email marketing, payments, and any industry-specific system of record.
- Where leads must arrive, per channel, and in what format — the integration that matters most and breaks most often.
- Analytics and search properties: GA4, GTM, and Search Console IDs with admin status, and whether access will be granted so historical data survives the rebuild.
- Domain registrar, DNS host, and email host with the account HOLDER named — never a password. Flag any account held by a former vendor, which is a migration risk and sometimes a hostage situation.
- Google Business Profile admin intent for each location, including whether access is delegated to seogmb@ciwebgroup.com.
- Must-keep URLs and the full redirect map for anything changing, with the old and new path per row.
- Any embed, portal, or third-party script that has to survive the rebuild, and what breaks if it does not.
- Accessibility and performance constraints the audience imposes, and any compliance requirement that touches the technical build.

15. **Compliance & legal** — license display rules, warranty/guarantee language operations can honor, Privacy / Terms URLs or “use CIWG template”, testimonials permission, SMS/TCPA if texting.

- Every disclosure this industry requires on the page, quoted exactly as it must appear. Approximations create liability.
- Licence display rules: which numbers must be shown, where, and in what format.
- Guarantee and warranty language that operations will actually honor, with the boundary conditions written out. A guarantee the field cannot deliver is a complaint generator.
- Privacy Policy and Terms: existing URLs, or an explicit instruction to use the CI Web Group template, plus any industry-specific addendum.
- Consent and permission status for testimonials, photos, names, and any customer-identifying detail.
- SMS/TCPA posture if the business texts: consent language, opt-out handling, and A2P 10DLC registration status.
- Regulated claims that cannot be made at all in this industry, listed so no agent generates one.
- Who signs off on published claims, by name or role. An unnamed approver means nothing gets approved.

16. **Competitors & whitespace** — 3–5 competitors with what to beat; local whitespace the new site should own.

- Three to five named competitors with URLs, and for each: what they do better, what they do worse, and the single thing to beat them on.
- Above-the-fold audit of each competitor: the offer, the price transparency, the CTA, the proof, and the guarantee they lead with. This is the fastest source of real whitespace.
- What no competitor in this market offers — pricing transparency, a real estimator, published availability, a named expert, a specific guarantee. That gap is the site's opening.
- Which competitor currently wins the money phrases, and with what page type. The format that wins is usually more instructive than the domain that wins.
- Any competitor consolidation, acquisition, or PE roll-up in the market, and how this business intends to sound like the alternative to it.

17. **Success metrics** — the outcomes the site is accountable for in the first 90 days.

- The two or three outcomes the site owns in 90 days, each with a number and a source of truth for measuring it.
- The baseline for each metric as it stands today, including how it is currently measured and how trustworthy that measurement is.
- Lead volume vs lead quality targets stated separately. Optimizing the first and reporting the second is how rebuilds get called failures.
- The revenue or margin logic behind each target, at whatever level of detail the operator will share.
- What would make the operator call this rebuild a success at 90 days, in their own words — and what would make them call it a failure.

18. **Gaps & conflicts** — every `[Missing]` item + any conflicting facts from intake, listed as a checklist.

- A checklist of every `[Missing — operator to fill during upload]` item, grouped by the section it belongs to, with who owns supplying it.
- Every `[Skipped by operator]` item, kept visible so a later reader knows it was a decision and not an oversight.
- Every conflict, in a table: `Fact | Source A says | Source B says | Which is provisionally used | Who resolves it`.
- The completeness score, and specifically which missing items are blocking the build versus merely thinning it.
- Anything the operator asserted that could not be verified anywhere, marked as unverified so no agent treats it as established fact.

19. **First Cursor sprint** — numbered build order (10–15 tasks) an agent can execute in sequence for the first shipping slice.

- Ten to fifteen tasks in dependency order, each small enough to finish and verify, and each naming the sections of this brief it reads from.
- The first task is always the shape of the thing: routes and IA from Site map, before any page is styled.
- A verification line per task — what must be true for it to count as done.
- Every task that is blocked by a `[Missing]` item, marked as blocked with the item named, so the sprint does not stall silently.
- What is explicitly out of scope for this first slice, so the sprint ends instead of expanding.

## Writing rules for the brief

- **Length: there is no upper limit worth respecting. Target 8,000–25,000 words** and go longer when the operator's inputs justify it. This is a knowledgebase seed, not a summary. If the brief is under 8,000 words, you have not finished interviewing.
- **Required means minimum.** Every section below lists what it must contain. Add H3 subsections, tables, and rows whenever the operator gives you more — never trim to fit a template.
- **Be specific and local.** A sentence a competitor could publish verbatim is a wasted sentence. Names, numbers, streets, hours, brands, model numbers, jurisdictions.
- **Never invent.** No prices, licences, certifications, rebates, warranties, review counts, or service areas that did not come from the operator. `[Missing]` is always the correct answer when you do not know.
- **Preserve the operator's words** where they are distinctive. Quote them and attribute them.
- **Tables where structure helps** — locations, catalog, personas, phrases, site map, redirects, conflicts. Prose where judgement matters.
- **Mark every unknown inline** with `[Missing — operator to fill during upload]` or `[Skipped by operator]`, in place, not only in the gaps section.
- **No secrets.** Account holders and property IDs yes; passwords, keys, and recovery codes never.
- **Never publish a cash price, an insurance-coverage claim, or a safe drive-away time the operator did not state explicitly.** In this industry the out-the-door price and the "when can I drive it" answer are the facts customers check first — and the safe-drive-away time is a safety commitment, not a marketing line.

---

## Asset library — collect alongside the brief

The brief is the text layer. These are the files the build needs, and collecting
them during the interview is far cheaper than chasing them mid-build. List what
exists, what is missing, and who owns getting it.

- Logo originals in vector, plus every variant in use.
- Brand guide or style sheet, if one exists.
- Real photography: team, premises, work in progress, finished work, equipment, product.
- Video: any existing footage, testimonials, walkthroughs, or ads.
- Documents customers receive: proposals, agreements, warranties, care instructions, price sheets.
- Credential artifacts: licences, certificates, insurance certificates, award notices.
- Existing content worth keeping: blog posts, guides, FAQs, spec sheets.
- Analytics exports: top pages, top queries, conversion paths, call recordings summary.
- Review exports from every platform.
- Adhesive spec sheets and the safe-drive-away time chart operations follows.
- ADAS calibration equipment list and a sample pre/post-scan calibration report with customer data redacted.
- Warranty documents and sample invoices with pricing redacted if needed.
- Before/after and damage photography with permission status per set.

---

## Quality bar — self-check before you emit

Do not output the document until every one of these is true. If one is false, go
back to the operator with the specific gap.

1. Every one of the 19 sections is present, in order, with the exact heading name.
2. No section is a single line. Any section that would be one line is a section you did not interview for.
3. Every location has a Google-verified address and a GBP URL, or an explicit `[Missing]` per field.
4. Every claim that could be checked has its proof named next to it.
5. No price, licence, credential, rebate, or service area appears that the operator did not supply.
6. The money phrases, the AEO questions, and the page modules agree with each other — a phrase with no page and a page with no phrase are both defects.
7. Every conflict between sources is recorded rather than resolved by guess.
8. No password, key, or recovery code appears anywhere.
9. The brief is at least 8,000 words, or the operator has explicitly said the remaining sections are skipped.
10. A competent stranger could build this auto glass replacement & repair site from this document without another conversation.

---

## Execution command

Acknowledge these instructions, request the intake data, and wait. Once the
operator provides it, interview them in rounds until every section is answered or
skipped — there is no question limit — then output **only** the single fenced
block for `HBIS-Cursor-Site-Brief.md`.

The operator will download that one file and feed it to Cursor for site design,
development, and optimization — and/or paste it into the CI Web Group AI-to-AI
upload step, where it becomes part of the Hydra knowledgebase.

---

## Version

- **HBIS Cursor Site Brief v2.0 — Auto Glass Replacement & Repair Edition** · 2026-08-05
- v2.0 replaces v1.0, which specified each of the 19 sections in a single line and capped the brief at 1,500–4,000 words. That cap was the reason briefs came back thin. There is now no upper limit, every section carries a full field-level specification, and this edition adds 10 industry knowledge domains and 7 industry-specific intake items.
- Generated from `scripts/hbis-prompt-core.mjs` + `scripts/hbis-prompt-industries.mjs` by `scripts/generate-hbis-prompts.mjs`. **Do not hand-edit this file** — edit the source and regenerate. `hbis-prompts-generated.test.mjs` fails if they diverge.
- The 19-section contract is validated by `lib/hbis-cursor-brief.ts`; section 7 is the XOR pair (`Services & products` | `Practice areas & services`) and this edition uses **Services & products**.
- Sibling editions and the searchable index: `/hbis`.
