Lookelo original research — September 20, 2026
Author: Oliver Zeman / Lookelo

STUDY A: 18 NEW ENGLAND BUSINESS HOMEPAGES
We fixed a purposive sample of three plumbing/HVAC businesses in each of the six New England states before measuring the pages. Discovery used public searches for plumbing businesses around Boston, Providence, Hartford, Portland, Manchester and Burlington, followed by named-business searches and service-area checks. Businesses outside a city can serve that market. This is a convenience sample, not a random sample or a list of top-ranking businesses. No population estimates, confidence intervals or causal claims are justified.

The sample is website-sample.csv. All 18 returned usable HTTP 200 homepage responses on September 20. Collection timestamps, final URLs and SHA-256 hashes are in website-observations.csv and website-observations.json. We fetched robots.txt first, respected rules for LookeloResearch/1.0, and did not replace failed observations. Sixteen robots files returned usable text; two returned 404. The protocol was recorded before measurement in docs/research/2026-09-20-plan.md in the website repository.

Measurements use downloaded HTML, not a rendered browser DOM. Title means text in the head/title element; descriptions must be nonempty; canonical counts measure link elements, not correctness of the target. H1 counts include repeated or hidden markup. A question heading means an H1–H6 containing a question mark, not an editorial judgment about answer quality. A tel link is an anchor whose href starts with tel:. Forms are any HTML form, including search or newsletter forms; we did not test submissions or classify them as lead forms.

JSON-LD means a script with application/ld+json that parses as JSON. Types are collected recursively. The local-business subset requires the exact type LocalBusiness, Plumber or HomeAndConstructionBusiness. This is not full Schema.org validation, coverage of all possible local subtypes, microdata or RDFa, or a test of rich-result eligibility. One parseable block is enough to count a page. Nine of the fourteen pages with JSON-LD had one of those local types.

Explicit bot groups mean an exact named User-agent line for Googlebot, OAI-SearchBot, GPTBot, PerplexityBot, ClaudeBot or Claude-SearchBot. The three AI-group observations each named at least one of the five AI bots. Presence does not mean permission. Absence does not mean blocking: wildcard rules can apply. These counts do not evaluate each bot's effective permissions, WAF behavior or ability to retrieve the page. Noindex checks cover robots/googlebot meta directives and response X-Robots-Tag headers; absence does not prove indexability or indexing.

No Google positions, search traffic, AI answers, AI citations, conversion rates, browser performance or customer outcomes were measured for these businesses. Raw responses are retained privately with hashes; the public dataset contains observations, not republished pages. The title extractor was corrected to exclude SVG title elements before publication, and stored responses were reprocessed. No findings were selected by whether a business looked good or bad.

STUDY B: FIVE PROVIDERS' PUBLIC PLANS
We inspected official pricing and documentation for OtterlyAI, Peec AI, Ahrefs, Semrush and Profound on September 20, 2026. These were selected as examples of different billing units and monitoring approaches, not a complete market survey. Tool observations and source URLs are in tool-observations.csv. Prices are USD as displayed; taxes, negotiated terms and future changes are excluded. Billing terms are stated per row. Peec's dynamically rendered page was verified in a browser. Profound's live page showed Trial and Enterprise; older search snippets were excluded. Semrush's captured pricing page explicitly stated billed annually; we do not present it as a verified month-to-month offer.

We did not buy subscriptions, execute vendor prompt panels, test mention accuracy, compare real brand scores or measure ROI. Vendor descriptions are claims about published plans, not independently tested performance. Our original analysis normalizes units and calculates example workloads. A modeled month is 30 days. Ahrefs' documented one-prompt/one-model/one-location unit is used only for its arithmetic, not assumed for every vendor. Other engine weights, repeat runs or extra locations can consume more capacity. Unused capacity and failures are not adjusted in the example. Provider capabilities require confirmation at checkout.

REPRODUCTION
The website repository contains scripts/research/audit_homepages.py and scripts/research/summarize_audit.py. Run the latter to regenerate summary.json from stored observations. Alternatively download calculate.py and website-observations.json from this research directory into the same folder and run python3 calculate.py. A fresh crawl can differ because sites change. Research data: /research/2026-09-20/website-observations.csv; /research/2026-09-20/tool-observations.csv; /research/2026-09-20/summary.json.

This is a Lookelo-authored observational study and desk audit. Lookelo sells website and search services. Findings do not establish that purchasing our services or any named tool improves rankings or AI citations.