Skip to content

Content depth for retrieval: expand the FAQ and thicken the thin pages #77

Description

@robbyrob42

Follow-on to #76, which fixed everything in the 2026-09-04 GEO audit that was code. What is left is copy, and it is the half that actually moves citations.

Where the measurements landed

Site-wide average 43.25/100, every page in the tool's foundation band. After #76 the structural categories should clear; these two do not, because no amount of markup fixes them:

RAG chunk readiness: 51/100. 19 sections, but only 4 fall in the 100–150 word range retrieval systems favour, and the average section is 42.3 words. Answer engines retrieve passages, not pages. A 42-word section is a fragment that gets pulled without enough context to stand as an answer, so it either gets skipped or gets cited misleadingly.

Thin pages. /experts is 124 words, /scope is 210. Both scored 37/100, the lowest on the property. /experts is the page attorneys land on when they search a name, and it is currently the thinnest thing on the site.

Also flagged and worth fixing in copy rather than markup: "few numerical data" on every page, and "no links to external sources" on every page. Both are E-E-A-T signals — concrete numbers and citations to authority are what separate a claim from an assertion, for a reader and a retrieval system alike.

1. Expand the FAQ

The deck carries five questions and #76 now emits all of them as FAQPage on /engagements. They are good questions — they answer objections. What is missing is the informational half: the questions counsel types into a search box before they have an objection, which is exactly what an answer engine is built to serve.

Every addition has to be answered in visible copy, not markup only. That is Google's condition for FAQ eligibility and it is also the honest position.

Draft candidates, grouped by what they do. These are prompts, not copy — the answers are yours to write.

Differentiators already decided that nothing on the site currently states:

  • "Do we engage you directly, or through a referral service?" — This is the no-third-party-billing decision made 2026-09-04, and it is a genuine differentiator worth publishing. Root System invoices retaining counsel directly. No broker sits between the engagement and the money, which means no markup on the published rate and no third party whose solvency counsel has to think about. TASA's collapse this year left experts unpaid; the practice took a position on that.
  • "Who actually does the work?" — The bench model. A dozen people assembled per matter, not a solo expert subcontracting out. The /experts ItemList markup now says who is published; the copy should say how staffing works.
  • "What does the $2,500 assessment produce?" — The rate card names it; nothing describes the deliverable. This is the lowest-commitment entry point on the property and the most under-explained.

Screening questions counsel asks every expert:

  • "How do you screen conflicts?" — Conflicts are screened before anything else and /scope says so, but the mechanism is not described. Note this intersects Conflict check: screen the expert’s investment holdings on the intro call #64, which tracks adding the expert's own investment holdings to the intro-call screen. Do not publish an answer that describes a step that is not yet performed.
  • "What jurisdictions do you take?"Blocked on a decision, and it is the one worth making. areaServed is deliberately omitted from the markup because the copy nowhere states a footprint. Under the directory-led plan that was a cheap trade. Under a content and answer-engine strategy, jurisdiction is a primary retrieval facet — "software expert witness in California" needs something to match on. The copy decision comes first; the markup follows it in one line.
  • "What do you need from us to start?" — Documents, access, timeline. Practical, and the kind of thing that gets retrieved verbatim.
  • "How quickly can you start?" — Availability is a screening axis. One business day to reply is published; time to engagement is not.
  • "Do you work plaintiff or defense side?" — May be deliberately unstated. Your call.

Requires your judgment, flagged rather than drafted:

  • "Have you ever been excluded under Daubert or Frye?" — Publishing an answer is a litigation-strategy decision, not a marketing one. Raised because opposing counsel will ask it regardless, not because it should go on the site.

2. Thicken the thin pages

/experts (124 words). The index is a list of one person with a one-line intro. It should carry the bench argument itself: why a dispute over what a system did crosses discipline boundaries, what the twelve people cover, how staffing against an approach differs from assignment off a rota. Most of that argument already exists in the bench block on /method — the index currently sends the reader elsewhere to find out what the page is about.

/scope (210 words). A form with a heading. What belongs around it: what happens after submission, what the conflict screen checks, what "not a secure channel" means in practice, and what the reader gets back and when. Every one of those reduces intake friction and gives the page something to be retrieved for.

Section length across the property. Where a 42-word section is doing real work, leave it. Where several short sections are making one argument in sequence, merging them into a 100–150 word passage makes the argument retrievable as a unit instead of as fragments.

3. Numbers and citations

Both flagged on every page.

Numbers: twenty-five years, a dozen experts, five years in the fab, $5.5M, thirty-five people, ten hours, $2,500, $450/hr, one business day, 90-day deletion. Most of these already exist somewhere in the deck. The finding is that they are not on the pages where the claims they support are made.

External citations: FRE 702 and its 2023 amendment, Daubert, the IEEE/SEMI publication, Cal. Civ. Code § 1798.140 (already cited in the privacy policy). Linking to authority is an E-E-A-T signal and it is also just correct practice for a page arguing about admissibility standards.

Acceptance

  • FAQ expanded, with every marked-up answer visible on the page.
  • /experts and /scope both carrying enough copy to stand as answers.
  • At least one measured number and one external citation on every substantive page.
  • Re-run uvx --from geo-optimizer-skill geo audit --sitemap https://forensics.rootsystem.com/sitemap-index.xml and compare; a baseline snapshot from 2026-09-04 is saved locally for geo diff.

Out of scope

The blog. Tracked separately.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions