Linker review: a score is a diagnostic, and the live audit failed
An honest source and live-app review: useful diagnostics, an audit that failed, simulated citations, and no verified MIT license.
LINKER · GEO / SEO · 25 SEP 2026
A site can have a tidy title and still never show up in an AI-generated answer. Linker tries to make that gap visible. It pairs SEO and GEO scores with a visibility checker, page comparison, content suggestions and reports. That is a good organizing idea for editorial work. But a score is a diagnostic, not proof that ChatGPT will cite you.
Reported facts
- On September 25 I read public commit 7a302aa and tried the live site. The source contains a Next.js audit API, scoring functions, a visibility workflow and comparison UI. Its README advertises 15 features and calls the project MIT-licensed. Those are project claims. There is no LICENSE file in the checked tree, and GitHub reports no license. The code is visible, but permission to reuse it is not verified. I did not install it or create an account. GitHub.
- I entered https://example.com on the public site. A staged progress view appeared, then the app returned to its landing page with “Failed to fetch.” I got no score, PDF or report. The progress sequence is interface animation, not evidence that each background stage completed. One failed attempt does not prove all audits fail; it does mean this review cannot claim a successful end-to-end result. Live app.
Reported facts
On September 25 I read public commit 7a302aa and tried the live site. The source contains a Next.js audit API, scoring functions, a visibility workflow and comparison UI. Its README advertises 15 features and calls the project MIT-licensed. Those are project claims. There is no LICENSE file in the checked tree, and GitHub reports no license. The code is visible, but permission to reuse it is not verified. I did not install it or create an account. GitHub.
A live test, not a success screenshot
I entered https://example.com on the public site. A staged progress view appeared, then the app returned to its landing page with “Failed to fetch.” I got no score, PDF or report. The progress sequence is interface animation, not evidence that each background stage completed. One failed attempt does not prove all audits fail; it does mean this review cannot claim a successful end-to-end result.
How one URL turns into two scores
The server fetches HTML and extracts title, text, headings, links and structured-data markup. Its heuristic grades six GEO factors: content clarity, structured data, authority, AI readability, citation readiness and entity coverage. It also grades six SEO factors: title, description, heading structure, keyword density, internal links and speed hints. The heuristic overall score weights GEO at 60% and SEO at 40%. If an AI-provider response arrives and parses, it may replace heuristic scores and findings; otherwise the heuristic result stays. These are page-readiness estimates, not a measured probability of citation. The “speed” factor uses markup hints rather than a timed browser test. Scoring source · Audit API.
Three ways to use it
- Publisher preflight: scan a public article before release and inspect its title, description, headings and structured data. Treat a low score as a question to investigate, not a command to add boilerplate. This is a plausible workflow from source, not a success I reproduced on the live site.
- Agency comparison: compare two to five public landing-page URLs, then inspect underlying content and link differences. The numbers are tool estimates, not Google rankings or observed AI-assistant answers.
- Editorial questions: use query themes to ask whether a page answers actual reader questions clearly and with attribution. Validate demand and citations independently. In my live test, query generation briefly reported one working model, then returned to its empty state with no visible completed list.
Where the visibility checker stops
The “AI Visibility Checker” does not query live ChatGPT, Claude, Perplexity or Gemini. Its source sends the URL and a candidate question to NVIDIA-hosted models, instructing them to simulate an AI search answer. It interprets the returned JSON as “mentioned” or “cited.” That may inspire wording, but it is not measured placement, a citation log or a four-engine comparison. Simulated results need to be labeled as such. Visibility source.
Analysis (Jack Gross)
My analysis: the strongest part of Linker is the checklist idea, putting page hygiene next to answer-focused writing and showing its scoring dimensions in source. The weak point is treating the numbers as ground truth about AI search. My live audit failed, visibility is estimated rather than observed, and some UI counters are generated demo values, not usage evidence. The missing license leaves reuse rights unclear. A security review is needed before relying on accounts or private client data; sensitive details and reproduction steps are deliberately left out of this public review.
Verdict
I would use the idea to guide a manual review. I would not quote the visibility score as a real answer-engine result, and I would not enter private client data into this beta. A real license, clear labels on what is simulated and a reproducible successful live audit would make a follow-up worth writing.
Live app · Public repository · Scoring source · Audit API · Visibility source. Commit 7a302aa, checked 2026-09-25.