AI transparency
How Connekt uses AI
AI in Connekt suggests and summarizes. People decide. This page explains what the AI does, what it sees, where it falls short, and how your team should review what it suggests.
Updated September 27, 2026. This is product documentation, not legal advice. Hiring laws differ by location and are changing quickly — check your own obligations with employment counsel.
In short
- Advisory only. The AI never rejects, advances, hides or filters anyone on its own. Every application stays in front of a person.
- Reasons, not just numbers. Each suggested fit comes with a short written reason, shown right next to it, so you can check it against the resume yourself.
- Newest first, not score first. Applicants are listed in the order they applied. You can sort by suggested fit for a moment, but it's never the saved default.
- Applicants are told. Your application pages show a notice about AI use before anyone applies, with a contact for accommodations or a review without AI.
What the AI is for
Connekt uses AI in three places, each to save you reading time, not to make a call for you:
- Suggested fit for applicants. When someone applies through your application link, the AI reads their resume (and cover letter, if they sent one) against the role you wrote, and suggests a fit from 0 to 100 with a 2–4 sentence reason. It's labelled "Suggested fit (advisory)" everywhere it appears.
- Finding people to contact. When you describe a role in plain English, the AI turns it into search criteria and writes a one-line reason for why each result matched. The order of search results comes from which of your criteria each profile matched, not from an AI judgement of the person.
- Sorting replies. When someone replies to your outreach, the AI tags the reply as interested, neutral or not interested, so you can follow up with the right people first.
The intended use is a small in-house HR team reviewing applicants and outreach for a role it defined, with a person reading each application before deciding anything.
What it must not be used for
Using Connekt's AI in these ways is outside what it's built and documented for. Please don't:
- Reject, advance or filter people on the score alone — for example only reading the top ten, setting a cut-off, or declining everyone below a number.
- Treat the score as the main or deciding factor, or use it to overrule what a person on your team concluded from the application.
- Screen for health, disability or personality disorders, or use any clinical or psychological test alongside it.
- Analyze emotion, voice, video or faces. Connekt doesn't do this, and it must not be added through uploaded material.
- Infer or act on protected characteristics — race, sex, age, disability, religion, national origin, pregnancy and so on — or proxies for them such as zip code, graduation year or photos.
- Make decisions about current employees such as pay, promotion, discipline or termination.
What the AI sees
For a suggested fit, the AI is sent only:
- the role title and description you wrote;
- the text of the applicant's resume;
- the text of their cover letter, if they included one.
It does not see the applicant's photo or any image, their email address or phone number, any demographic information, or anything about other applicants. It is instructed to judge only job-related skills and experience and to ignore names, addresses, zip codes, dates of birth, graduation years and gaps explained by health or caregiving. Instructions reduce the risk; they can't remove it, which is why a person reviews every application.
For search, the AI sees your role brief and public profile fields (title, skills, headline). For replies, it sees the text of the reply.
The model behind it
Connekt uses Claude, a general-purpose foundation model hosted by Anthropic and called through its commercial API. Connekt did not build or train this model. It is not fine-tuned or trained on your data or your applicants' data: each request carries only the information listed above, and the model is given fixed instructions that Connekt writes and versions.
Because the underlying model was trained by Anthropic on broad public and licensed text, Connekt can't list its training data in detail. What Connekt controls is the instructions, what's sent, and how the output is shown — and those are what this page describes.
Limitations
- It only knows what's on the page. A strong candidate with a thin or unusual resume can get a low suggested fit.
- It depends on your role description. A vague description gives vague suggestions. Describe the actual work and the must-have skills.
- It can be wrong, and confidently so. The reason may mention something the resume doesn't say, or miss something it does. Always check.
- It can carry bias. Language models learn from human writing and can reflect its patterns. Connekt hasn't yet run an independent bias audit of the suggested fit.
- Formatting matters. Scanned or image-only PDFs have no readable text; those applications show as failed rather than getting a guess.
- Numbers aren't comparable across roles. A 70 for one role doesn't mean the same as a 70 for another.
How to review AI suggestions
A few habits keep the decision with your team, where it belongs:
- Open and read every application you decide on — the resume, not just the number.
- Read the written reason and check it against the resume. If it's wrong, ignore it.
- Weigh the suggestion alongside everything else you know; it's one input, not the answer.
- Make sure whoever decides can override the suggestion, and knows what it can and can't see.
- Record your decision by setting the applicant's status. Connekt logs who changed it and when.
- Answer accommodation and "review me without AI" requests promptly. They go to the contact shown in your application page's AI notice, which you can change in Settings.
- If an applicant asks how AI was involved in their outcome, you can point them to this page.
Records we keep
For every suggested fit, Connekt records which model and which version of its instructions produced it, which kinds of information it was given, a summary of the result, and when. For every status change on an applicant, it records who made it, from what, to what, and when. These records are kept for at least four years, which is what California's rules on automated hiring tools ask of employers, so your team can answer questions or support an audit later.
Changes
When the model or its instructions change in a way that could change results, we update the version below and let customers know.
September 27, 2026 ·
applicant-fit-2026-09-27Applicant scores relabelled "Suggested fit (advisory)" with the reasoning always shown. The scoring instructions now tell the model to judge only job-related skills and experience and to ignore protected characteristics and proxies for them. Every score and every status change is now logged.
Questions about how Connekt uses AI? Contact us. This page is not legal advice.