The Generation Effect:
Why Writing Your Own Sermon
Changes Your Brain in Ways AI Cannot.

There is a reason you remember the sermon you wrestled with all week but forget the article you skimmed this morning. Cognitive science has a name for it. It is called the generation effect. And it is one of the most powerful arguments against letting AI write your sermons.

86 Studies. One Conclusion.

Bertsch and colleagues published a meta-analysis covering 86 studies on the generation effect. The finding: information you generate yourself — as opposed to information you merely read or receive — is remembered better by an effect size of .40. That is almost half a standard deviation.

In plain language: when you wrestle with a passage, form your own outline, draw your own connections before consulting any source, the material embeds into your memory at a fundamentally deeper level than if someone hands you the answer.

This is not marginal. In educational research, an effect size of .40 is considered meaningful enough to shape policy. It is the difference between a student who remembers a concept three months later and one who forgets it in three days.

Now apply that to sermon preparation. A pastor who generates his own exegetical insights, forms his own theological connections, and builds his own application points from scratch is encoding that material into long-term memory. He is being formed by the text. The text is shaping him while he shapes the sermon.

A pastor who reads an AI-generated outline and edits it is consuming, not generating. The generation effect does not activate. The formation does not occur.

What Happens Inside Your Brain

The generation effect is not a metaphor. It is a neurological event.

When you actively generate information — working through a text, making connections, creating structure — three brain regions activate simultaneously: the prefrontal cortex (executive reasoning), the inferior temporal gyrus (semantic processing), and the parahippocampal gyrus (memory formation). This triple activation creates stronger, more durable neural pathways than passive reading.

Donald Hebb coined the principle that neuroscience still operates on: “use it or lose it.” Only synaptic connections exercised through consistent use are preserved. Inactive connections deteriorate and may vanish. The brain refines motor neuron pathways with practice but reduces them without use.

Arthur and Bennett's meta-analysis of skill decay across 189 data points from 53 articles confirmed the pattern: cognitive, accuracy-based tasks are the most susceptible to decay. That describes theological reasoning precisely. And the decay ranges from near-zero immediately after training to d = -1.4 after 365+ days of nonuse.

Every week a pastor skips the generative struggle of personal exegesis, the neural pathways for theological reasoning weaken. Not dramatically in any single week. But cumulatively, over months and years, the capacity erodes.

The London Taxi Driver Study

The most striking visual evidence for what active cognitive struggle does to the brain comes from London.

Woollett and Maguire followed 79 aspiring London taxi drivers for four years with MRI scans. To earn a license, these drivers must memorize 25,000 streets within 10 kilometers of Charing Cross plus thousands of landmarks — a process called “The Knowledge” that takes three to four years of intensive study.

At baseline, there were no differences in hippocampal volume between those who would eventually qualify and those who would fail. Zero. Same starting brain structure.

Four years later, every single qualified driver showed visible growth in posterior hippocampal grey matter. Failed candidates showed no change. The act of active navigation — struggling, getting lost, finding the way — physically restructured their brains.

“The struggle is not a bug to be optimized away — it is the mechanism of formation.”

A separate study compared taxi drivers (who navigate freely) with bus drivers (who follow fixed routes). Despite similar years of driving experience, taxi drivers had larger posterior hippocampi. It was not driving that changed the brain. It was active navigation — the struggle of finding one's own way.

And the inverse: GPS users showed a correlation of r = -0.68 between GPS hours and spatial memory decline. The more they relied on the tool, the more their own capacity degraded. “Those who used GPS more did not do so because they felt they had a poor sense of direction” — ruling out reverse causation.

The Sermon Prep Parallel

The parallel is not subtle. A pastor who navigates the text freely — struggling through the Greek, wrestling with the context, getting lost in the argument before finding the way — is building theological hippocampal grey matter. A pastor who follows the AI's fixed route is a bus driver.

The Turkish math study (PNAS 2025) proved this in an educational context. Roughly 1,000 students were divided into three groups: control, unrestricted GPT, and GPT with tutoring guardrails. During practice with AI, the unrestricted group looked 48% better than control. But when AI was removed for the exam, they scored 17% worse.

The mechanism was clear: students used unrestricted GPT as a crutch — copying solutions instead of generating them. The generation effect never activated. They never built the underlying understanding.

A pastor who copies AI-generated sermon outlines is using a crutch. And when the crutch is removed — when the passage is difficult, when the cultural context is unfamiliar, when the congregation is hurting and needs a word from someone who has been in the text all week — the 17% deficit shows up as theological shallowness.

Irenaeus Knew This

The generation effect is not new. The church has always known that formation happens in struggle.

The early church father Irenaeus argued that God could have created humanity in a state of perfection but chose instead to create beings who would grow through struggle toward maturity. As the philosopher John Hick summarized Irenaeus: “Hard-won virtues are more valuable than ready-made perfections.”

That is the generation effect stated as theology. The struggle of sermon preparation is not an inefficiency to be eliminated. It is the process by which the preacher is formed. When you wrestle with a text until it breaks you open, you are not just producing a sermon. You are being produced. The text is doing its work on you before you do your work on the congregation.

AI cannot replicate that process because AI cannot struggle. It produces output. It does not wrestle. It generates tokens. It does not encounter God. The pastor who skips the struggle in favor of AI efficiency is not saving time. He is forfeiting formation.

The Cognitive Reserve Argument

There is a long-term dimension the church should find sobering. Research on cognitive reserve — the brain's ability to resist neurological decline — shows that people with less than 8 years of education have a 2.2x higher dementia risk. Low lifetime occupational cognitive engagement raises dementia risk by 2.25x. Higher cognitive reserve decreases dementia risk by 46%.

Active cognitive work — the kind demanded by rigorous theological study — builds the reserves that protect the brain across a lifetime. Every week a pastor spends in deep generative study is not just preparation for Sunday. It is investment in a mind that will serve the church for decades.

Every week a pastor skips that study in favor of AI-generated content is a withdrawal from that account.

The Right Role for AI

None of this means AI has no place in sermon prep. It means AI has a specific place — after the generation has occurred.

Generate first. Wrestle with the text. Form your own outline. Draw your own connections. Let the Spirit speak before any tool speaks. Then use AI to check your cross-references, surface historical context you missed, verify your original-language analysis. Let AI confirm and expand what you already generated.

That is the sequence that preserves the generation effect. That is the workflow where AI sharpens your study instead of replacing it.

OpenLumin was built for this exact role. It retrieves evidence from 15+ scholarly sources — commentaries, Ancient Near East cultural context, original-language data, cross-references, Theographic Bible data covering 3,000+ people, places, and events. Every claim is sourced. Every citation is marked as verified or flagged for review.

It does not generate your sermon. It equips the pastor who already did the work.

Because the sermon was never just the product. The formation that created it was always the point.

About: AI Fluency Ministry helps the church understand and use AI wisely. OpenLumin is the practical application of that research — a free Bible research companion that retrieves evidence so you can do the thinking.