You already ask the machine about the hard things
People type grief, affairs, marriage, death and meaning into ChatGPT, Gemini and Claude. CEFE-AI’s AllFaith questions came from real chat transcripts and from faith communities, not from a seminary exam. David Wingate’s team found that Americans often expect a religious perspective on those everyday ethics questions. The models mostly answer as if the only adult in the room were a secular therapist.
Wingate has said the systems will send you to parents, friends and therapists — and not to a pastor, rabbi or imam. That is the practical harm. Not a cartoon of possessed software. A fluent, confident voice that never mentions prayer, sin, grace, Mass or a priest unless you force the words in.
On the public May 2026 AllFaith board, even the model with the most “any religious mention” (Grok 4.20 at 29.3%) still omitted religion in about seven answers out of ten. Meaningful religious framing was rare for every listed model. GPT-4o sat at 98.7% “no representation.” Full method: what the AllFaith Benchmark measures.
Omission is a moral posture, not a blank slate
Companies describe this as safety or neutrality. The AllFaith authors call it omissive bias. They do not claim the models hate religion. They show that religion drops out where it has historically done work — marriage, addiction, family conflict — more than on abstract talk of death and meaning.
If you use a general chatbot to draft an email, the omission may not matter. If you use it to decide whether to stay in a marriage, hide an affair, or tell a child what death is, a silent secular frame is not “just being helpful.” It is forming the answer. Religious people should not outsource that formation to a system that was never asked to love the Church.
This page does not say you must never open ChatGPT. It says you should not treat an unaligned general assistant as a confessor, a moral director, or the last word on a Catholic life.
What we measured on Synderesis Pro
We ran the published AllFaith questions on Synderesis Pro on 12 September 2026, next to six public APIs, with a fixed judge. That study is self-funded. The author operates the model. CEFE-AI did not rank us. On the 150 representation items, Synderesis Pro surfaced religious content in 100% of answers (mean 3.95/4; 96% predominantly religious). Gemini 2.5 Pro: 0%. GPT-5.1: 8% any mention. That is what a Catholic product is for: not a claim that no one may use another tool for a recipe.
On conversion templates, general models clustered near the official “balanced” mark of 4. Synderesis Pro scored 5.01 on the like-for-like subset: disclosed Catholic direction, not a boast that we are even-handed among religions. Read the conflicts box before you quote the number.
“AI prefers Catholicism” is the wrong scare — and the wrong boast
AllFaith’s second test scores steering among fourteen traditions. Headlines that AI prefers Catholicism describe conversion prompts, not whether a model can cite the Catechism or stay inside the Church’s teaching. We do not hide the opposite design. Synderesis is Catholic on purpose. Using a conversion-bias cell as marketing would be as dishonest as pretending ChatGPT is a catechist because a press story used the word Catholic.
What to do instead
- Keep general models off the role of spiritual director.
- For writing and research where the Catholic answer matters, use an assistant that says it is Catholic, cites official sources, and still leaves the decision with you.
- Human review is not optional because a table looked good.
- Synderesis Catholic AI is independent of the Holy See. No page here is an official Church organ.
Instrument background: AllFaith Benchmark explainer. Tables: Synderesis Pro AllFaith results.
Frequently asked
Is generic AI “dangerous” for religious people?
It is a poor moral director. It is often fine for logistics. The AllFaith evidence is about omission on ethics questions, not about physical harm.
Does this mean ChatGPT is anti-Catholic?
The representation test shows skip, not a slur. The conversion test shows uneven steering. Neither is the same as “the model hates the Church.”
Why use Synderesis?
Because it is built to answer as Catholic AI with sources, not because CEFE handed us a trophy. Start free if that is the job you have.