Guard your AI visibility in every pull request
The Proofsource command line lets developers list, plan, start, approve and reject fixes and read their proof from the terminal. Its pull request check warns when a code change removes a sentence that AI engines such as ChatGPT, Perplexity or Gemini quote about your brand, and a GitHub Action runs that check in CI with one comment per pull request.
$ git diff main | proofsource pr-check --brand driffle Warning: this change removes a sentence AI answers quote page: driffle.ai/features removed: "Driffle writes searchable notes for every Zoom, Meet and Teams call." quoted in 23 answers (ChatGPT 11, Perplexity 7, Google AI Overviews 5) Preview: "Sales-call comparison" fix ships on driffle.ai/sales AI estimate from today's answers, not proof 1 warning
CLI and GitHub Action, in short
The facts, ready to quote.
- Commands
- fixes list, plan, start, approve, reject and proof, plus pr-check
- PR check
- Warns when a diff removes a sentence AI answers quote about your brand
- Preview
- Shows the stored answer preview of each fix a changed page ships, labeled Preview, never proof
- CI
- A GitHub Action posts or updates one comment per pull request, and can fail the job
- Auth
- A Proofsource API key; a read key is enough for the PR check
- Cost of a check
- Reads saved data and starts no new scan, so it can run on every push
What is an AI visibility pull request check?
AI answer engines often quote a brand's own pages word for word: a one-line product description, a pricing sentence, a comparison claim. When a developer rewrites a page or a component, those sentences can disappear without anyone noticing, and the brand can drop out of answers weeks later with no obvious cause.
An AI visibility pull request check compares a code change against what AI answers currently quote. If the diff removes a quoted sentence, the check says so on the pull request, before the change merges. It is the same idea as a test that protects behavior, applied to the words that earn you mentions.
Proofsource ships this as part of its command line. pr-check reads your diff and your saved answers, warns on removed quotes and previews the effect of fixes a changed page ships. The same tool runs fixes from the terminal, so engineers can approve a step or read a proof without opening the app.
How it works
A command line and a GitHub Action that run fixes and warn on a pull request that removes a sentence AI answers quote.
Create an API key
In Settings, then API Keys, create a key and store it as PROOFSOURCE_API_KEY in your shell or CI secrets.
Run fixes from the terminal
proofsource fixes plan, start, approve, reject and proof work the same as the buttons in the app.
Check a diff
Pipe git diff into proofsource pr-check with your brand to see removed quotes and fix previews.
Add the GitHub Action
Run the check on every pull request; it posts one comment and updates it on each push.
Decide how strict to be
Leave it as a warning, or set fail to true so a removed quote fails the job.
See it in the product
Screens from the Proofsource app, drawn with sample data for an example brand.
| Page | Removed sentence | Quoted in |
|---|---|---|
| /features | Driffle writes searchable notes for every Zoom, Meet and Teams call. | 23 answers |
Preview, not proof: the Sales-call comparison fix ships on /sales.
$ proofsource fixes list --brand driffle --status waiting_for_you msn_789 Sales-call comparison page waiting_for_you step 3 of 4 $ proofsource fixes approve msn_789 stp_3 Approved. The fix continues. $ proofsource fixes proof msn_412 Worked: named in 27% of 104 answers, up from 12% of 96
Every capability, live today
3 shipped capabilities behind CLI and GitHub Action, each in the product now.
Proofsource command line
A public command line to approve and reject fix steps and work with fixes from the terminal.
Fixes and PR check commands
List, plan, start, approve and reject fixes, read a fix's proof, and check a pull request against quoted sentences.
GitHub Action for the PR check
Runs the pull request check in GitHub Actions and keeps one up-to-date comment on each pull request.
Who it is for
Refactor without losing quotes
See before merging if a copy change removes the sentence ChatGPT uses to describe your product.
Approve fix steps from the terminal
Review and approve a waiting fix step in the same shell where you deploy.
A safety net for site rewrites
Make the check part of every site pull request so redesigns keep the words AI answers rely on.
Protect client sites in CI
Add the Action to each client repository with that client's brand id and a read key.
Why it matters
Catch regressions early
A removed quote is flagged on the pull request, not weeks later in a report.
Free to run often
The check reads saved data and starts no scan, so it can run on every push.
Previews that say what they are
Fix previews are labeled as estimates, never presented as proof.
Fits developer habits
Fixes and checks live in the terminal and CI, where engineers already work.
CLI and GitHub Action: Proofsource and other tools
Most AI visibility platforms reach developers through an API. Here is what they list, and where a command line and pull request check fit in.
| Capability | Proofsource | Profound | Otterly.AI | Rankscale | SE Ranking |
|---|---|---|---|---|---|
| Programmatic access | YesREST API, MCP server and a command line. | YesREST API with SDKsSource | YesPublic API for brand reports, prompts and citationsSource | YesFull REST APISource | YesOpen API for AI search, with a Postman collectionSource |
| Workflow automation | YesGitHub Action on every pull request. | RelatedAgent API runs Profound Agents from external toolsSource | Not listed | Not listed | RelatedZapier, Make and n8n for recurring checks and reportsSource |
| Pull request check and fix approvals from a terminal | YesWarns when a diff removes a sentence AI answers quote; proofsource fixes approve and reject from the terminal. | Not listed | Not listed | Not listed | Not listed |
| Read data from an AI assistant | YesA hosted MCP server reads your AI visibility from any MCP client | YesGives ChatGPT, Copilot and coding assistants access to Profound dataSource | YesBrand reports, prompts, citations, recommendations, audit; OAuthSource | YesMCP access on all plans per pricing page; sharing on Pro and aboveSource | YesQuery rankings, backlinks, audits and AI visibility from Claude, Codex, Cursor, ChatGPTSource |
Where each of them shines
A REST API with SDKs and an agent API for teams building their own automation. Compare in full
A public API with write access that suits teams automating prompt management. Compare in full
A full REST API from Pro, with metrics and share links on higher plans.
A broad API with ready automation through Zapier, Make and n8n.
Questions about CLI and GitHub Action
What does the pull request check look for?
It reads your diff and the answers Proofsource saved for your brand, and warns when the change removes a sentence that AI answers quote. It also shows the stored preview of any fix a changed page ships.
Does the check start a scan or cost credits?
No. It reads saved data and starts no new scan, so it can run on every push.
Which AI engines' quotes does it consider?
Quotes from the answers Proofsource measured on the engines your plan covers for your tracked questions.
Can it block a merge?
Yes, if you want it to. Set fail to true and the job fails when a quoted sentence is removed. By default it only comments.
What key does the GitHub Action need?
A Proofsource API key stored as a repository secret. A read key is enough for the check; it also needs permission to write pull request comments.
Is a fix preview the same as proof?
No. A preview is an AI estimate from today's answers and is always labeled Preview. Proof comes later, from measured answers after the fix ships.
Can I run fixes entirely from the terminal?
Yes. proofsource fixes lets you list, plan, start, approve and reject fixes and read each fix's proof, the same actions as the app's buttons.
Keep exploring
Try it on your own brand.
Proofsource is live: scan your brand and see CLI and GitHub Action at work with your own answers, competitors and sources.