Kaizen AI Lab · Project SOAP
PahRoo Appraisal & Consultancy · www.pahroo.com
Search is splitting into two surfaces. The first is the familiar results page. The second is the answer an assistant gives when a prospect asks it a question instead of typing a query. This review assesses how ready www.pahroo.com is for the second surface, using checkpoints that were added to the Kaizen framework specifically to test it.
Ordered by severity. Each finding names the checkpoint it came from and the evidence behind it.
| Severity | Checkpoint | Finding and evidence |
|---|---|---|
| HIGH | GC-14 | Entity graph defects: anonymous duplicate nodes or conflicting property sets 29 business-entity node(s) across 16 pages; 3 anonymous (no @id); 1 distinct @id; 0 @id with conflicting property sets; name variants ['PahRoo', 'PahRoo Appraisal & Consultancy']; sameAs 4 |
| HIGH | GC-17 | Answerable facts are not stated as extractable declarative sentences absent across all sampled pages: ['years in business', 'turnaround time', 'pricing approach', 'hours']. present: ['credentials', 'service area'] |
| HIGH | GC-21 | Agent and assistive-technology readability gaps on key pages Measured through the rendered accessibility tree, not the raw markup. Homepage: 21 of 68 links carry no accessible name, including anchors whose only target is '#'. Contact page: 6 of 8 form fields have no programmatic label, relying on placeholder text alone, which disappears on input and is not a label. Both pages: no main landmark, so a machine reader cannot identify the primary content region. Homepage heading order skips levels five times. Three dropdown triggers expose no expanded state. |
| MEDIUM | GC-19 | Vague superlatives used where verifiable specifics belong 3 distinct superlative(s) across 3 page(s): /about-pahroo-appraisal-consultancy/: ['highly experienced']; /service/commercial-property-appraisal-services/: ['unparalleled']; /philadelphia-real-estate-appraisals/: ['industry-leading'] |
| MEDIUM | GC-31 | Buyer decision structure has uncovered dimensions 7/10 covered on sampled pages. gaps: ['causes and consequences', 'costs and tradeoffs', 'selection criteria'] |
| MEDIUM | GC-34 | E-E-A-T artifacts are partial: authorship exists in schema only, with no visible byline, no author credentials, and no editorial accountability statement Re-verified after the first probe hit HTTP 429 and measured an error page. Present on all three articles tested: author schema naming Michael Hobbs, datePublished (2025-06-18 to 2026-07-27), and 13 to 40 outbound reference links. Absent on all three: any visible byline in the rendered page, appraisal credentials attached to the author, and any editorial policy, 'reviewed by', or fact-check statement anywhere on the property. |
Recorded because they are load-bearing. A property that fails any of these cannot be helped by content or authority work until it is fixed.
| Checkpoint | Result and evidence |
|---|---|
| GC-01 | No snippet suppression on sampled money pages 16 pages checked for meta nosnippet, max-snippet:0, data-nosnippet, and X-Robots-Tag |
| GC-03 | Robots posture is not contradicted at the edge for sampled agents 6 agent classes fetched the homepage: all HTTP 200, no cf-mitigated header, response sizes within 50% of the human baseline (103,196 bytes) |
| GC-03b | Training-crawler policy is explicit named in robots.txt: ['GPTBot', 'ClaudeBot', 'CCBot', 'Google-Extended', 'Applebot-Extended']. This is the client's business decision (GF-03); surfaced, not changed. |
| GC-16 | No placeholder or template values in JSON-LD An initial broad scan matched the string 'placeholder' 7 times; all were verified as HTML form placeholder attributes and one CSS ::placeholder selector. No template values such as yourdomain.com, YOUR STREET ADDRESS, or XXXXX are present on the sampled pages. Note: the 2026-07-28 baseline found genuine placeholder schema on the Naples and Phoenix pages; those pages now carry real values. |
| GC-20 | No self-serving AggregateRating on own business schema third-party review velocity remains the correct lever (GF-11) |
| GC-32 | Comparison coverage present 5 comparison-shaped URL(s) in 155 sitemap URLs: ['/zestimate-vs-appraisal/', '/rural-vs-urban-appraisals/', '/how-appraisers-compare-homes/', '/retrospective-vs-current-market-value-appraisal/', '/divorce-home-appraisal-vs-price-opinion/'] |
| GC-33 | Property publishes non-reproducible information found: ['original research or survey', 'named framework'] |
| GC-39 | Specification-level facts are extractable from HTML One PDF link exists on the property, a hud.gov handbook, which is an outbound government citation rather than PahRoo facts locked in a document. It is not linked from any sampled money page. No page combines heavy HTML with thin extractable text, so no JS-only fact surfaces were found. |
Each recommendation carries two labels. The first states where the guidance comes from. The second states how strong the evidence is that it works, on a scale where E1 is a documented prerequisite, E2 is well supported by multiple independent lines of evidence, and E3 is directionally supported with causality unresolved. Kaizen publishes both so that proven work and reasonable bets can be told apart.
| Ref | Recommendation | Impact | Effort | Horizon | Owner |
|---|---|---|---|---|---|
| GF-13 |
Give every link and form field a programmatic name
Twenty-one homepage links and six of eight contact-form fields are currently unnamed in the accessibility tree. Assistants that read a page through that tree cannot describe or follow an unnamed control. Adding real labels and a main landmark also fixes the same defect for screen-reader users, which is why this is the least contestable item on the list.
|
High | Low | Immediate | Kaizen implementation |
| GF-05 |
Consolidate the business entity to a single identified node
Twenty-nine business-entity nodes appear across sixteen pages. Three carry no identifier at all, so they read as separate businesses, and the name appears in two forms. Assistants need one unambiguous record to name a firm with confidence. This is recommended because accuracy is independently correct, and the causal evidence that consistency alone increases citations is not settled.
|
High | Low | 30 days | Kaizen implementation |
| GF-08 |
State the facts buyers ask for as plain declarative sentences
Four facts a prospect always asks are absent from the property in extractable form: how long the firm has practised, typical turnaround, how fees are determined, and office hours. Assistants lift clean declarative statements almost verbatim and do not lift facts buried in marketing prose. This improves how cleanly a page can be used once it has been retrieved. It is not a claim that the change causes the page to be found, and it ships in staged batches with untreated pages held for comparison so any retrieval effect is measured rather than assumed.
|
High | Medium | 30 to 60 days, staged | Kaizen drafts, firm approves |
| GF-22 |
Make authorship and editorial accountability visible
Articles name an author inside structured data only, where no reader sees it. There is no visible byline, no appraisal designation attached to the author, and no editorial standards statement anywhere on the property. Expertise that is asserted achieves nothing; expertise that exists as an artifact can be assessed.
|
Medium | Low | 30 days | Kaizen implementation, firm confirms credentials |
| GF-19 |
Close the three uncovered stages of the buyer decision
Seven of ten decision stages are covered. Three are not: what drives a valuation up or down, what an appraisal costs and the tradeoffs involved, and how to choose an appraiser. Assistants decompose a buyer's question into several narrower searches, and those searches reach every stage of the decision. Coverage means a small number of genuinely useful pages, one per real topic. It does not mean a page per search phrase, which is a pattern search platforms treat as abuse.
|
High | Medium | 60 to 90 days | Kaizen drafts, firm approves |
| GF-10 |
Replace remaining superlatives with verifiable specifics
Three pages describe the firm with unverifiable adjectives. A specific, checkable number is quotable by an assistant and an adjective is not. Same staging discipline applies as the declarative-facts work.
|
Low | Low | 30 days, staged | Kaizen drafts, firm verifies each specific |
| GF-03 |
Decide the training-crawler posture deliberately
The property currently names five training and model-development crawlers in its crawler directives, and search-oriented crawlers are reaching the site without obstruction at the network edge. Whether to permit model training is a business decision, not a technical one, and it is presented here as a choice rather than changed.
|
Medium | Low | Business decision | PahRoo decides |
Several tactics sold widely in this market do not survive the evidence, and Kaizen does not bill for them. Search platforms state that specialised AI files and markup carry no ranking effect, that content does not need artificial chunking, and that no special writing style is required. The FAQ rich result was retired in May 2026, so question-and-answer markup earns its place by making content extractable and not by producing a search feature. Building a knowledge graph, publishing self-authored ranking lists, and generating one thin page per search phrase are all either unsupported or actively risky.
Fourteen further checkpoints in the framework could not be assessed because they depend on data sources Kaizen does not yet have access to for this property. Each is listed with the access required and the person who owns restoring it. An unassessed checkpoint is recorded as unavailable. It is never recorded as a pass, and no visibility conclusion is drawn from a check that did not run.
| Checkpoint | What it measures | Access required | Owner |
|---|---|---|---|
| GC-02 | Search generative AI feature state (Include / Exclude / Inherit) | Search Console access | Carson Vasquez |
| GC-04 | Brand Radar report identifier for custom prompt tracking | Ahrefs interface, human creation step | Jennifer Villadolid |
| GC-05 | Frozen money-question set | Client onboarding step | Jennifer Villadolid |
| GC-06 | Verbatim assistant probes: mentioned, cited, recommended, per surface | Contained browser tier | Carson Vasquez |
| GC-07 | Brand mentions, history, and share of voice | Ahrefs route | Carson Vasquez |
| GC-08 | Impressions across brand name variants | Ahrefs route | Carson Vasquez |
| GC-10 | Cited domains and cited pages for the category | Ahrefs route | Carson Vasquez |
| GC-11 | Cited domains intersected with competitor referring domains | Ahrefs route | Carson Vasquez |
| GC-15 | Fact reconciliation against Google Business Profile at value level | Business Profile quota approval | Carson Vasquez |
| GC-22 | Generative AI performance impressions | Search Console access | Carson Vasquez |
| GC-23 | Assistant referral sessions separated from automated traffic | Analytics access | Carson Vasquez |
| GC-24 | Crawler hit volumes by operator | Server or CDN log access | Carson Vasquez |
| GC-35 | Probe sampling sufficient to handle answer variability | Contained browser tier | Carson Vasquez |
| GC-37 | Representation accuracy: whether assistants describe the firm correctly | Contained browser tier | Carson Vasquez |
Stated so the results can be judged on their evidence.