A score from 0 to 100, with a traffic light per area
Green from 75, amber from 50, red below. At a glance it is clear whether the Google profile, the website or the reviews hold the business back.
RepoCity RadarAuditor
RadarAuditor measures how visible a business is on Google, on the map, in directories and in AI assistants — and what is missing.
For roofers, dentists, kitchen studios and every business whose customers come from nearby. The first audit runs from the outside and needs no access to your accounts.
What the customer gets
Three things, all from measurements of the business itself and of two or three competitors in the same town.
Findings in plain language, with a ready fix
Each finding says what was measured, what it costs and what to do. When the fix is a text or a piece of code, it comes ready to copy.
A forecast as a range, explicitly an estimate
How many more enquiries are possible is shown as a range from low to high, with every assumption listed underneath. It is not a promise.
low estimatehigh estimate
- Clicks on position 1: 25 to 35 of 100 searchers, German average
- Enquiries in the trades: 5 to 10 of 100 visitors
The images show the app with sample data.
How an audit runs
A job goes round once. Scoring follows fixed rules; writing happens only at the end. Depending on the size of the website it takes a few minutes to half an hour.
- App on the phoneThe operator enters the business with its location, two or three competitors and the search terms it earns money with, then starts the remote audit.
- SupabaseThe job lands in the database queue. If a step fails, it is retried up to three times.
- Collector containerFetches the website, load time, Google profile, positions in Google and Bing search, directories and the answers of ChatGPT, Gemini and Perplexity.
- ScoringEach of the 104 checks is passed, partly passed, failed or not measurable. Code computes the points, not a language model.
- Claude Code on the PCReads the raw data and check results and writes one finding with a fix per open point, plus the report texts. Before anything is uploaded, the collector validates the result against a fixed data format.
- Report as PDFThe collector assembles the report as a print-ready file. It appears in the app and can be walked through with the customer on a tablet.
Twelve areas, one number
The score is a weighted average of twelve areas. Weights, checks and thresholds live in a single check catalogue; the weights add up to 100.
- Technical foundation of the websiteWeight 10
- Page structure and meta informationWeight 10
- Content and demand coverageWeight 12
- Company data on the websiteWeight 6
- Structured dataWeight 8
- Google Business ProfileWeight 14
“Do you have your Google profile in your own hands?”
- ReviewsWeight 8
- Directories and data consistencyWeight 7
“Can people find you on the iPhone in Apple Maps?”
- Links and authorityWeight 6
- Rankings and visibility in searchWeight 10
- Visibility in AI assistantsWeight 5
“Does ChatGPT name you when someone asks for a business like yours?”
- Law and trust in GermanyWeight 4
A thirteenth area, impact with customer access (calls, clicks and enquiries via Google), is only measured with access and does not count towards the number. The questions are translated from the German check catalogue.
What is measured, and what is not
A lot can be measured from the outside, but not everything. What is missing appears in the report as “not measurable” and does not count towards the points, instead of being guessed.
Measurable from the outside, no access needed
- Encryption, load time on phone and PC, layout on the phone
- Page titles, headings, texts, broken links
- Name, address and phone on the website compared with Google, Apple Maps, Bing and business directories
- Machine-readable company data for Google and AI assistants
- Google profile: categories, opening hours, photos, number and average of reviews
- Positions in Google and Bing search and on the map, compared with the competitors
- Whether ChatGPT, Gemini and Perplexity name the business, and whether what they say is correct
- Legal notice and privacy policy
Only with customer access or paid data
- Calls, route requests, clicks: figures from the Google profile, Search Console and Analytics are only available with access.
- Replies to reviews: the public interface does not deliver them; an audit with access reads them from the profile.
- New reviews per month: Google only returns the five newest publicly, so the figure is an estimate.
- Description, services, products, posts, questions and answers, service areas and attributes of the Google profile: no interface; the operator checks them by eye.
- Visibility around the location, links from other sites, search volume: only with a paid data service, which is switched off today.
- Positions in Google search: searching in the browser can hit a block page; then positions are missing for that audit.
Built as a product, not as a demo
| Python tests | 983 passing |
|---|---|
| App tests | 81 passing |
| Check catalogue | 104 checks in one file, the single source for weights and thresholds |
| App | Installable web app for phone, tablet and PC |
| Collector | Docker container, runs on the PC and later unchanged in the cloud |
| Data | Supabase: database, sign-in, file storage and queue, hosted in Frankfurt |
| Delivery | Cloudflare Pages |
| Texts | Claude Code on the subscription, no usage-based costs |
| Invoices | Mandatory details under section 14 of the German Value Added Tax Act (Umsatzsteuergesetz), gapless numbers, unchangeable once issued |
| Tax | Year folder with DATEV export for the tax office |
Status: built; the first run against a real business is still to come.
Who it is for
Businesses
The remote audit needs nothing from you: no access, no password, no appointment. For ongoing care you grant access to your Google profile once and, if you use it, to Search Console. Following the guide this takes about five minutes, and your password stays with you. After that a report arrives every month showing what has changed.
Companies and recruiters
One check catalogue is the single source for checks, weights and thresholds; code, app and texts derive from it. Code computes the points by fixed rules; the language model only writes the texts and takes every number from the measured data. The hand-off to Claude Code is validated against the data model, and invalid results stay behind. Issued invoices are protected against changes in the database.
Built by Daniel Hütte, AI engineer from Münster, looking for a position as AI Agent Engineer.