The short version
In this paper
| How a run works | The path from your request to a stored result |
| What is checked, and how it is scored | Rules, gating, business-type overrides, the four results, the score bands |
| Where the data comes from | Eight sources and how each is read |
| How AI is used, and where it is not | What language models do, what they never touch, the controls |
| What you receive | Result page, PDF, 90-day memo, actions, share link, monthly re-run |
| Your data and security | Sign-in, storage, the crawler, plan enforcement, hosting |
| Reliability | Status page, retries, monitoring, human review |
How a run works
When you request a score, the web app records the job and hands it to a separate worker. The worker fetches your site, asks each source its questions, evaluates the 86 rules, stores every result with its evidence, and writes the memo. Because audits never run inside the web app, a slow website or a slow provider cannot freeze your dashboard, and a run that fails halfway is retried rather than lost.
What is checked, and how it is scored
The engine is a rule interpreter: every number on your report can be traced to a rule, a source and a piece of evidence.
The rules. 86 rules in 10 weighted categories: website foundation, technical SEO, local presence, reviews and reputation, social media, conversion capture, content and email, analytics and ads, brand consistency, AI visibility. Each rule states what it checks, its weight, its severity, the source that answers it and the fix it recommends. The full list is public in the knowledge base.
Gating. Three checks matter more than the rest: the site serves over HTTPS, the business profile on the leading map and search platform exists and is claimed, and a visitor can actually reach you (phone, form or email). If any of these fails, the overall score is capped at 40, because nothing else works well until they are fixed.
Your type of business. Restaurants, home services, B2B services, ecommerce and medical or dental practices each switch certain rules on, off or up in weight. A rule about a menu does not count against a law firm.
Four results. A rule returns pass (counts fully), partial (counts half), fail (counts zero) or not assessable. Not assessable means a source was unavailable or the check could not run; the rule is left out of the score with the reason shown. It is never counted as a failure, and it never inflates a score either: the report shows how many rules were assessed in each category.
The score. A category score is the weighted pass rate of its assessable rules; the overall score is the weighted mean of the categories, then capped by gating. Bands: 90–100 excellent, 75–89 good, 40–74 needs work, 0–39 poor. The free score runs a subset of the rules (website, DNS, page speed, search) and labels the rest as locked.
Evidence. Every result keeps what was checked: the query sent, the tag found, the count, the page inspected, the provider that answered. You can hand a finding to whoever fixes it without re-explaining it.
Where the data comes from
Every source is read through an adapter that records the call, so a missing credential becomes “not assessable”, never a wrong answer.
| Source | What it tells us | Good to know |
|---|---|---|
| Your website, read by our crawler (EchoRankBot) | Page content, titles, links, forms, structured data, image alt text, footer year, placeholder text | Obeys robots.txt, identifies itself, one request per second, at most 60 pages, never leaves your domain |
| DNS records | Email authentication: SPF, DKIM (common selectors), DMARC | Read live at the time of the run |
| Page speed test | Loading speed and layout stability on mobile | Industry-standard lab measurement from the leading search platform |
| Places listing | Whether your listing exists, its rating, review count, hours, name-address-phone consistency | Public listing data only |
| Business profile, if you connect it | Owner-side details, posts and review replies | Unlocks 7 more checks; you can disconnect at any time |
| SEO data application | Whether your site is indexed, where it ranks for your brand and for “service + city”, and whether it appears in the local 3-pack | No webmaster account needed, so the free score works before you sign up |
| AI answers | Whether the major AI assistants name your business for “best <service> in <city>” and describe it accurately | Our sister product, EchoRank |
| Language model | Reads copy for a few judgement calls; writes the memo | See the next section |
How AI is used, and where it is not
Language models write and classify. They never decide a score.
Used for
- Two judgement calls about copy: whether your homepage says what you do, for whom and where in its first 100 words; and whether an AI-generated description of your business matches your site.
- Writing the 90-day memo: turning the ordered list of failed rules into at most eight plain-language actions with steps and an owner. The order itself is fixed by the rules: gating first, then critical, high, medium, low.
- On the Growth and Managed plans, drafting posts, review replies and briefs from your evidence and your own brand voice, always held for your approval.
Not used for
- Scoring, weighting, gating, or the pass, partial or fail decision on any rule.
- The figures in your PDF: the “why” line under each top fix is a fixed template over stored evidence, so the same run always produces the same report.
- Anything a visitor sees before signing in.
Controls
- Every generated text passes a claim check before it can be approved: no guarantees, no competitor mentions, no unsupported numbers. Anything that fails goes to a human reviewer.
- Prompts are versioned files; each call is logged with its prompt version, tokens and cost, and each account has a monthly cap.
- Nothing generated is ever published to one of your channels without an explicit approval from you.
What you receive
| Deliverable | Built from | How |
|---|---|---|
| Result page | The run, every rule result and its evidence | Tabs for audit, evidence, recommendations, roadmap and progress; free results show locked categories openly |
| PDF report | The same run | Generated on demand from the stored results, never edited by hand; A4, page numbers, the no-guarantee note once at the end |
| 90-day memo (paid plans) | Failed rules ordered by gating, severity and category | At most eight actions, each with its rule ids, an owner (you, or EchoPilot drafting for your approval), an effort size and a target week; quick wins under an hour are separated out |
| Open actions | The memo | Tracked on your dashboard; the next memo takes what you completed into account |
| Share link | A run | Revocable; free results expire after seven days |
| Monthly re-run (paid plans) | Every rule, again | The delta per category is stored, so progress is measured rather than claimed |
Your data and security
Signing in. Sign in with your existing account at the leading search platform, or with an emailed link that works once and expires quickly. Sign-in attempts are rate-limited. Passwords are never stored because none exist.
What we store. Your business details, each run’s rule results and evidence, memos and actions, and billing state. Fetched web pages are not kept beyond the run. Tokens for connected services such as your business profile are encrypted at rest. Deleting your account removes your businesses, runs, memos and tokens.
What we do not store. Anything you paste into the free tools; page bodies; card numbers (our payment processor holds those); free-score data older than seven days.
How the crawler behaves on your site. It obeys robots.txt, names itself with a link to a page describing it, sends one request per second, reads at most 60 pages with a 10-second limit each, and never follows links off your domain. It cannot be pointed at private networks.
Plans are enforced on the server. What your plan includes is decided by one rule set on our side, checked on every action and every scheduled job; nothing depends on what a page happens to show.
Where it runs. On a dedicated server in the United States behind an edge network, with encrypted connections end to end. The service is not offered in the EU, EEA or UK.
Reliability
- A public status page shows whether the database, the queue, the worker and each data source are healthy right now.
- Retries, not silent failures. Audit and memo jobs retry on transient errors; a source that stays down produces “not assessable”, with the reason on your report.
- Every provider call is logged with its cost, so usage is monitored per account and spikes are caught the same day.
- A human in the loop. Blocked drafts, judge flags, scraper failures and sudden rating drops land in an operator queue that is reviewed daily.
As of 28 September 2026, rule set v0.1. Content drafting on the Growth and Managed plans is rolling out. EchoPilot improves the marketing signals you can control — your website, listings, reviews, content and online presence. No honest marketing platform can guarantee rankings or revenue, because competitors and the platforms themselves affect those outcomes too. © 2026 ChatLogic Insights Ltd, registered in England and Wales, no. 15593166.