# Instacart Names 3 Screening AIs. Opting Out of Tofu Isn't Enough

_Instacart's AI screening notice names Metaview, Tofu and Real Talent. What each one scores or flags, when to opt out, and how to ask why you were rejected._

- Canonical URL: https://www.refolk.ai/candidates/blog/instacart-ai-screening-notice-opt-out-explanation
- Published: 2026-11-01
- Reading time: 10 min
- Keywords: Instacart AI screening notice, how to find out if AI rejected my application, request human review job application AI, Connecticut SB 5 AI hiring notice, Metaview ideal candidate profile ranking

If you apply to Instacart, software scores your application, summarizes it, and checks whether you're a real person, all before a human decides anything. Instacart's AI-Assisted Screening Tools Notice names that software and explains how to route around part of it, but the obvious opt-out leaves two fraud checks running.

Most employers never say this much, so read the notice closely before you click Apply.

## What Instacart's AI screening notice actually discloses

Instacart's AI screening notice names three tools: Metaview, Greenhouse Real Talent and Tofu. Metaview summarizes and ranks you against a profile of the ideal hire and also runs fraud detection. Real Talent and Tofu exist to flag fraud. The notice also tells you how to request review without Tofu's fraud check and how to ask for an explanation after a rejection.

The three tools, from the notice and the vendors' own material:

| Tool | What it does to your application | What it reads |
|---|---|---|
| Metaview (plugs into Greenhouse) | Summarizes, scores and ranks you against an "Ideal Candidate Profile"; also runs fraud and automation detection | Your application and the job description |
| Greenhouse Real Talent | Flags possible fraud or misrepresentation | IP address, email address, phone number |
| Tofu | Screens for fraud and identity misrepresentation | IP addresses, devices, document metadata, location consistency |

The notice also names IPQualityScore as a data processor. Put together, your network connection and contact details are being analyzed for fraud alongside the resume itself.

## Metaview ranks you against an Ideal Candidate Profile, not against a person

An Ideal Candidate Profile (ICP) is a description of the target hire that Metaview generates from the job description plus whatever extra context the recruiter adds, and then uses to evaluate every applicant. Your score measures how closely you resemble that profile. It doesn't measure how good you are.

Metaview's Application Review became generally available on March 17, 2026. Each applicant gets an ICP Fit rating of **Great, Good, or Okay**, with a written rationale. Metaview's own help docs are frank about it: matches "reflect alignment with the Ideal Candidate Profile you approve, not a judgment about candidate quality."

What that means in practice:

- **The job description is the source document.** If the posting says "experimentation platform" and your resume says "A/B testing infrastructure," a human sees the same skill. A profile built from the posting's wording may not.
- **The rationale is what the recruiter reads.** A "Good" with a short explanation may be the first description of you that reaches a person.
- **The notice doesn't show you the ICP.** The words in the posting are the only stand-in you have.

### Why when you apply might matter as much as what you write

Metaview describes the ICP as self-learning: "as you make decisions to accept and reject, the AI learns and updates the ICP." If Instacart has it set up that way (the notice doesn't say), a later applicant is scored against a profile adjusted by the recruiter's calls on earlier candidates, not only against the original posting.

That's an inference from Metaview's documentation, not a confirmed detail of Instacart's setup. The practical advice is the same either way: apply early, while the profile is closest to the posting you're tailoring against.

## Three tools check for fraud, and opting out of one leaves two running

Instacart's notice lets you request review without Tofu's automated fraud screening, but Metaview's entry in the same notice also covers "application fraud and automation detection," and Real Talent has its own separate alternative-process right. If you opt out of Tofu only, you're still being fraud-checked twice.

Metaview turns fraud detection on by default, according to its own candidate-screening material. So all three vendors in the notice do some form of fraud or authenticity checking.

This matters because of the baseline. Tofu publishes data on how often legitimate applicants come out clean, and the numbers are blunt.

> **50.3%** Baseline clean rate across Tofu's full applicant population
>
> Applicants with accounts on everyday personal platforms scored clean at 82% to 95%.

That gap is the mechanism. Tofu treats a personal digital footprint on everyday platforms as a signal that you're human. With that footprint, applicants landed in the 82% to 95% band. Without it, you're closer to the full-population baseline, where only about half came out clean. Tofu also estimates that mid-level remote engineering jobs draw an average of 20% to 30% fake applications, so these systems are tuned to be suspicious, and a real person with an unusual profile pays for that.

## Who should request human review before applying

Request human review of your job application if anything in your digital trail makes you look unlike a typical applicant. Real Talent analyzes your IP, email and phone. Tofu looks at "IP addresses, devices, document metadata, and location consistency across the application lifecycle." Ordinary, honest situations can trip those checks.

Opt out if any of these apply to you:

1. **You created a new email address for your job search.** A fresh Gmail with no history can raise a flag.
2. **You apply through a VPN.** Your IP says one city and your resume says another. That's what location-spoofing detection is built to catch.
3. **You moved recently.** Your resume lists your new city, while your phone, older documents or past IP history point to the old one.
4. **You're changing careers.** If your new field doesn't show up anywhere in your footprint yet, the story can look invented.
5. **You keep a deliberately thin online presence.** Few personal accounts puts you nearer the 50.3% baseline than the 82% to 95% band.

The notice says opting out "will not negatively affect your consideration for the role." Take it at its word. The alternative is a fraud flag you'll never see, applied before any human gets to you.

That's the worst part of automated fraud screening: it fails silently. You don't get rejected for being unqualified, you just never get looked at, and from your side it feels exactly like every other application that disappeared. If you're job hunting while employed, the instinct to use a separate email and a VPN to stay discreet is the same thing that makes these systems suspicious of you.

## How to send the opt-out requests

Email CandidateQuestions@instacart.com, the contact the notice gives for the Tofu opt-out, before or right after you apply, and name each tool you want removed. Don't assume one request covers everything, because the notice gives Tofu and Real Talent separate rights.

1. **Apply with the details you'll actually use.** If you can, use a personal email with some history, your regular phone number, and no VPN.
2. **Email CandidateQuestions@instacart.com the same day.** Put the role title and requisition ID in the subject line.
3. **Name Tofu and Real Talent separately.** Ask for review without Tofu's automated fraud screening, and separately invoke the alternative process the notice gives for Real Talent. If Real Talent requests go somewhere else, ask them to forward it.
4. **Ask about Metaview's fraud detection.** The notice doesn't describe an opt-out for it. Asking costs nothing, and the reply tells you where things stand.
5. **Keep the thread.** If you're rejected later, you'll want proof of when you asked.

A short version that works:

> Subject: Alternative review request, [Role], [Req ID]
>
> I applied for [Role] on [date]. Per your AI-Assisted Screening Tools Notice, I request review of my application without Tofu's automated fraud screening, and I invoke the alternative process available for Greenhouse Real Talent. If Metaview's application fraud and automation detection can also be excluded, please do so. My contact details: [email, phone].

## Write so the summarizer repeats your strongest lines

Metaview writes the summary and fit rationale a recruiter sees first, so your best lines need to be ones a summarizer will pick up and repeat: plain, specific, and in the posting's own vocabulary. Clever phrasing tends to get paraphrased away, while concrete claims tied to the job description survive.

The recruiters reading these outputs rarely present themselves as tool specialists. In Refolk's index of professional profiles, pulled November 1, 2026:

| Segment | Count |
|---|---|
| US profiles with Recruiter or Talent Acquisition titles | 97,201 |
| ...that also list Greenhouse as a skill | 120 |
| UK profiles with the same titles listing Greenhouse | 2 |
| US vs. UK Greenhouse-tagged recruiters | 60x |

> **0.12%** US recruiters who list Greenhouse as a skill
>
> 120 of 97,201 profiles in Refolk's index. Skill tags are self-reported, so this isn't a count of companies using the tool.

The 120 who do list it cluster in New York and the SF Bay Area, at employers including Palo Alto Networks, Meta and Paraform. The point isn't that recruiters can't use Greenhouse. It's that the people reading your AI summary aren't, by and large, people who think of themselves as experts in the tool producing it. That summary is often the first, and possibly the only, description of you a recruiter sees.

How to write for it:

- **Match the posting's nouns.** The ICP is built from the job description. If it says "SQL" and "dbt," those exact words should appear next to your results.
- **Put one claim and one number in each line.** "Cut month-end close from 9 days to 5" survives summarization. "Drove transformational finance outcomes" gets dropped.
- **Front-load your top three bullets.** Summaries compress, and the first lines of each role are the ones most likely to make it through.
- **Keep employer, title and dates consistent.** Fraud detection checks consistency across your application, and a mismatch can read as misrepresentation.

Doing this by hand for every posting is slow, and it's the work [Refolk](/candidates) takes off you. Paste the posting, and I rewrite your own resume in its vocabulary without inventing anything you didn't do.

## The 30-day explanation request: who gets it and how to word it

If you're a candidate in Illinois, Colorado or Connecticut and Instacart rejects you, you can ask for an explanation within 30 days of the rejection, but only when the AI tools were a "material factor" in the decision. Outside those three states, the notice doesn't promise you anything, though asking politely still costs nothing.

The explanation covers three things:

1. The principal reasons for the decision.
2. The degree to which the tools' outputs contributed.
3. The categories of data used.

Item two is the closest you'll get to finding out whether AI rejected your application. Item three tells you whether your IP, email or phone played a role, which is the fraud path. Ask for all three by name.

> Subject: Explanation request, [Role], [Req ID], rejected [date]
>
> I received a rejection for [Role] on [date]. I am a [state] resident. Under your AI-Assisted Screening Tools Notice, I request (1) the principal reasons for the decision, (2) the degree to which Metaview, Tofu or Greenhouse Real Talent outputs contributed, and (3) the categories of data used. If an ICP Fit rating or fraud flag was generated, please include it.

Send it within a week of the rejection, not on day 29. If the answer points to a fraud signal and you never opted out, reply asking for a re-review through the alternative process. If it points to fit, the rationale tells you which words the profile was looking for.

## What Connecticut SB 5 does and doesn't give you

Connecticut SB 5 (Public Act 26-15) currently does one thing for rejected candidates: since October 1, 2026, an employer can't defend against a discrimination complaint by saying an automated decision tool made the call. The pre-decision notice duty applies only to deployments on or after October 1, 2027.

Instacart's notice covers Connecticut candidates, but it goes beyond what the law requires today. Don't expect the same from other employers because of SB 5:

- **Pre-decision notice:** not legally required in Connecticut until October 1, 2027.
- **Post-rejection explanation:** Connecticut has no equivalent requirement. Colorado does, requiring both pre-decision notices and post-adverse-outcome notifications within 30 days.
- **Discrimination complaints:** available now, and "the algorithm did it" isn't a defense.

Instacart's 30-day window follows Colorado's model. If you're in Connecticut, the explanation right you're using comes from Instacart's notice, not from SB 5.

## FAQ

### Does opting out of Tofu remove all fraud screening at Instacart?

No. The notice also lists Metaview for "application fraud and automation detection," and Metaview enables fraud detection by default. Greenhouse Real Talent has a separate alternative-process right. Request both opt-outs in writing, and ask whether Metaview's check can be excluded too.

### Can I see my Metaview ICP Fit rating?

Not directly. Metaview gives recruiters a Great, Good or Okay rating with a rationale. If you're in Illinois, Colorado or Connecticut and the tools were a material factor in your rejection, you can request the principal reasons and how much the tools contributed within 30 days, and ask for the rating in that request.

### Will requesting human review hurt my chances?

The notice says opting out "will not negatively affect your consideration for the role." Given Tofu's 50.3% baseline clean rate, the bigger risk for anyone with a new email, a VPN or a recent move is staying in the automated process and getting flagged without knowing it.

### Does Connecticut SB 5 require employers to explain AI rejections?

No. Connecticut has no post-rejection explanation requirement, and its pre-decision notice duty starts with deployments on or after October 1, 2027. Colorado requires post-adverse-outcome notifications within 30 days. What SB 5 gives you now is that an employer can't use an automated tool as a defense in a discrimination complaint.

---

_From the Refolk job search blog. More posts: https://www.refolk.ai/candidates/blog/index.md_