AI-Gutted: Companies Whose Product AI Absorbed, in the Numbers

Some companies die because customers leave. A newer pattern is stranger: the product itself gets absorbed. A general-purpose AI model starts doing the thing the company charged for, and the business hollows out while the website stays up. Our assessment has a name for these records: gutted. This piece walks through what the data shows about them, using only the assessment itself.

Of the 10,525 companies assessed in WillItSurvive's July 2026 assessment, a blind backtest of 116 dead or gutted companies produced negative verdicts on 90% of them and zero safe-to-join calls, using only facts knowable 18 months before each terminal event.

What "gutted" means in this assessment

A gutted company is not always a bankrupt one. Chegg still operates. Stack Overflow still operates. But the blind backtest treated them the way it treats terminal cases, because the mechanism is the same: the revenue-bearing product lost its reason to exist. In the backtest pairing file, several of these records appear on both sides of a pair, which is itself telling. Chegg (cutoff September 2024) was seeded as the "healthy survivor" twin for Course Hero, and Course Hero (cutoff January 2025) was seeded as the twin for Chegg. The blind judges, given masked, time-boxed facts, returned not called safe on both, in both replicates.

That was not a mistake. In the July 2026 backtest, 10 companies seeded as healthy survivor twins, including Chegg, Fisker, Nikola, and Getir, were flagged negative by the blind judges, and each turned out to be genuinely failed or gutted rather than healthy. The assessment caught all 10, in both replicates. Homework help is the clearest AI-absorption story in the set: two rival study platforms, each nominated as the other's healthy comparator, and neither one actually healthy by the time the twins were checked.

The AI-absorption cases in the pairing file

The backtest pairs include a cluster of companies whose product overlapped with what large models now do for free or nearly free:

Chegg and Course Hero (cutoffs 2024-09 and 2025-01): paid homework and study help, the category most directly answered by chatbots. Both drew not called safe in both blind replicates.
Stack Overflow (cutoff January 2026, paired against GitHub): community Q&A for programmers, the traffic source most visibly displaced by AI coding assistants. It sits in the dead arm of the backtest.
Kite (cutoff May 2021, paired against Tabnine): AI code autocomplete that arrived before the model wave and was overrun by it. The candidate list for its pair included GitHub Copilot, the product that redefined the category.
Jasper AI (cutoff September 2023, paired against Copy.ai): AI copywriting built as a wrapper on models that customers could increasingly use directly.
One Hour Translation / Blend (cutoff July 2024, paired against TransPerfect): human translation marketplaces, squeezed as machine translation quality closed the gap.
Udacity (cutoff September 2022, paired against Coursera) and Unacademy (cutoff January 2023, the seeded twin for Byju's): online education records under pressure, with Unacademy also among the 10 flagged twins.
EyeEm (cutoff October 2021, paired against Shutterstock): stock photography, a market generative image models reshaped.

The pattern across these pairs is that gutting spreads through categories, not companies. When the study-help twins both fail the blind check, and the AI-gadget twins both fail it, the assessment is telling you the pressure was on the product category itself.

Why the backtest matters for this pattern

It is easy to say, in 2026, that homework-help sites were exposed. The harder question is whether the exposure was visible in advance. The backtest was designed to answer exactly that: dead companies were judged on facts knowable 18 months before their terminal event, with identities masked, and matched surviving peers were judged at the same cutoff with the same masking.

In the July 2026 blind backtest, the dead arm of 116 companies drew 105 negative verdicts (90%), 11 middle verdicts (9%), and zero positive verdicts.

An identifiability probe checked this. Recognition-independence is claimed only for the dead arm, since the clean survivor subset (n=5) is too small to lean on.

On the survivor arm, two independent replicates agreed on the verdict bucket for 49 of 53 companies (92%) in the July 2026 backtest, so the judgments are stable rather than noise.

One publishing rule matters here. We never say "we predicted these deaths" and we never declare an outcome for a living company. The supported claim is narrower: given only pre-collapse facts, the assessment separated the doomed from matched survivors, and never once said safe-to-join on a company that was dead within 18 months.

What a reader can do with this

If you are evaluating a job offer or a vendor contract at a company whose product overlaps with general-purpose AI, the assessment suggests treating that overlap as a dated watch-item, not a verdict. Concrete things to check yourself:

Traffic and usage trends since the relevant cutoff. For the study-help pair, the blind cutoffs were September 2024 (Chegg) and January 2025 (Course Hero); verify what has happened since.
Whether revenue depends on a task a chatbot performs. Kite's category was absorbed by Copilot-class tools; ask whether the company's paid feature is a model wrapper.
Whether the "healthy competitor" is actually healthy. The twins readout shows rivals nominated as strong comparators can be gutted themselves; a weak category makes every peer look relatively fine.
Standing labels and answer cards. About 10,800 companies in the assessment carry a structured answer card with a WATCH section (assessed July 2026); use those dated items as a checklist, not a conclusion. Contested records, 1,571 of them in the July 2026 assessment, carry an independent second opinion.

Questions this data answers

Q: Which companies in the assessment were hollowed out by AI rather than by ordinary competition? The backtest pairing file includes Chegg, Course Hero, Stack Overflow, Kite, Jasper AI, Humane, Rabbit Inc., EyeEm, Udacity, and One Hour Translation/Blend, all in categories where general-purpose AI absorbed the paid task. Several, like Chegg and Stack Overflow, still operate but were judged terminal-pattern in the July 2026 blind backtest.

Q: Did the assessment actually distinguish gutted companies from healthy ones, or does it flag everything? In the July 2026 backtest, 90% of the 116-company dead arm drew negative verdicts with zero safe calls, while only 17% of 58 matched survivors drew negatives in the first replicate. That gap is the evidence the method separates rather than blankets.

Q: Both Chegg and Course Hero were flagged. How can rivals both be in trouble? Each was seeded as the other's healthy twin, and blind judges returned not called safe on both, in both replicates. When AI absorbs a product category, every company in it is exposed at once, so the usual "strong competitor" comparison stops working.

The probe checked. On the 17 dead companies the probe could not identify, 88% still drew negatives and none drew a safe call in the July 2026 backtest, so the dead-arm result does not depend on recognition.

Q: Does WillItSurvive say which currently operating companies AI is gutting next? No. Forward-looking material is published only as dated watch-items with a disclaimer, and we never declare an outcome for a living company. Readers get the dated facts and the checklist; the verification is theirs.

Assessments reflect evidence available July 2026. Companies change; treat watch-items as things to verify, not conclusions.