Ask us the hard questions

1By what standard do you judge?

Scripture — read in the historic Protestant confessional tradition. We are not neutral, and we are not going to pretend to be.

This is the most important answer on the page, so we will be blunt about it. That Scripture alone is the final rule for faith and practice is itself a confessional position. It is close to what Protestants and Roman Catholics have disagreed about for five hundred years. A service that measured everyone against it while claiming to stand nowhere would be doing something dishonest.

So we say it plainly: TruthRadar measures a sermon against Scripture, and it treats Scripture as the final authority. If you do not share that premise, you will not accept our conclusions, and you are entitled not to. We would rather you disagreed with our standard than misunderstood it.

What we will not do is pretend the alternative — a view from nowhere — exists. Every assessment of a sermon is made from somewhere. Most of them just do not tell you where.

2Who decided what counts as an error?

A person did. The standard was written and is owned by TruthRadar’s founder, who holds final responsibility for it, and it is published — every rule that decides a score, in plain English. You can read what counts as error, what never does, and argue with any of it.

What we will not claim is that a committee of scholars agreed it. No such committee exists. That is a real limitation and it belongs here rather than in the small print.

3What is an “in-house denominational debate”, and who drew the line?

This is the fairest hit on the page, and the honest answer today is that the rule exists and the boundary behind it is not yet published in enough detail.

The rule is that debates within Christian orthodoxy are never penalised. A sermon does not lose points for the tradition it is preached in — for example, for being Baptist rather than Presbyterian — nor for its view of baptism, or of church government, or of the end times.

But orthodoxy needs a boundary, and a promise without a boundary is not worth much. Wes Huff, whom we would take seriously on this, makes the difficulty vivid: he calls justification “probably the most important issue” between Protestants and Rome. So plainly not everyone agrees about what is in-house.

What we are doing about it: publishing what we treat as first-order, what we treat as genuinely disputed among orthodox Christians and never penalise, and what we treat as neither. Until that is published, this answer is an IOU, and we would rather label it as one than let it pass for a settled position.

4You score preaching from traditions that do not share your standard.

Yes. And we will tell you so rather than let you discover it.

There is a narrower thing worth saying. What the score assesses is whether the sermon handled the passage it preached faithfully — whether the text was represented, whether claims made from it are in it, whether the plain sense was respected. Preaching from a tradition that differs from ours can do that superbly, and much of it does.

Where our standard bites differently is on questions where the sufficiency of Scripture is itself the point at issue. Where that happens, the reasoning says so, rather than burying a confessional disagreement inside a number.

5Do you score Latter-day Saints, Christian Science or New Thought preaching?

No, and not because we are avoiding the question. We measure sermons that claim the historic Christian faith against the Scriptures as that faith has always read them. Those three traditions differ from it at the foundation — over who God is and who Christ is, the questions settled at Nicaea — so a score would not be measuring a sermon; it would be publishing a verdict on a religion. That is a different thing from what we do, and we would rather say so plainly than let a low number stand in for an argument we have not made.

Where a preacher of any tradition stands inside that historic faith and disagrees with us about how to read a passage, they are scored like everyone else, and the reasoning shows where the disagreement lies.

6Is a number on a sermon a category mistake?

Partly, yes — and that is why the number is never alone.

Faithfulness is not a quantity in the way length is. What the score does is compress a judgement into something comparable, which is useful for finding things and dangerous if taken for precision.

So the reasoning is always shown, always free, and quotes the sermon’s own words wherever there is a line to quote. We would rather you read the finding and disagreed with it than accepted the number.

And we publish our own imprecision. When we measured it in October 2026, across 294 sermons our current standard scored more than once, the highest and lowest results for the same sermon were typically about three tenths of a point apart, half a point on average, and in the worst case nearly four points apart. Where they are 0.8 or more apart, we show no score at all: the sermon is marked Not assessed. That was 52 of the 294. So treat two sermons within half a point of each other as the same, and anyone building a ranking out of tenths is misusing our own tool.

7Whose confession does the engine hold?

Ours, stated openly — see the first question above. The answer “none, it just reads the Bible” would be the least credible sentence on this page.

8A machine cannot discern spiritually.

Agreed. It cannot, and we do not claim it does.

TruthRadar does not tell you whether a preacher is godly, called, converted, or filled with the Spirit. It cannot see any of that and it does not try. What it does is narrower and mechanical: it reads what was said, compares it with the passage it came from, and reports where the two do not match — showing you the words it based that on.

That is a literacy task, not a spiritual one. It is the same task you perform when you look up a verse the preacher quoted and read the surrounding chapter. We are not claiming to do something you could not do. We are claiming to do it consistently, at a scale you have not got time for, and to show you the evidence so you can check us.

If 1 Corinthians 2:14 rules out our tool, it also rules out concordances, commentaries, study Bibles and every other instrument a Christian uses to read carefully. We think the objection proves too much.

9Whose theology did the model absorb?

We do not fully know, and neither does anyone else. That is worth stating rather than hiding.

A large language model is trained on an enormous quantity of human writing and carries whatever tendencies that leaves. We cannot open it up and audit its theology.

What we can do is constrain it and measure it. It works to the standard we publish, and it is given the sermon’s words and the Bible text — nothing else: not the preacher, not the church, not the tradition. And we test the output for bias.

We have found some, and the honest version of the story is that we found it twice. In early September we noticed our confidence measure — how sure the engine says it is of its own score — ran along denominational lines, and we stopped it influencing the order sermons appear in.

Days later we checked it properly, against the thing it claims to describe. Every sermon in the borderline range is scored three times, and we record how far those three results land apart. That spread is what disagreement actually looks like. Grouped by tradition, the spread is flat — between 0.34 and 0.45 everywhere. Confidence swings by 58 points across the same groups.

The sermon that settled it was not the one we expected. The three runs agree MORE on Black church preaching than on Baptist preaching, and the engine calls itself far less certain about it. A number that moves with tradition and not with the disagreement it is supposed to measure is not measuring anything.

So we stopped showing it. What appears on a sermon page now is how many times it was scored and how far those scores differed, which is a real measurement rather than the engine’s opinion of itself. The measure is still recorded, and if we can ever make it track disagreement instead of tradition it can come back.

That is the honest answer to this question. Not ‘the model is unbiased’, but ‘we look, we get it wrong, we look again, and we tell you what we found’.

That is the honest answer to this question. Not “the model is unbiased”, but “we look, and when we find something we say so and change it.”

10What happens when the model changes?

Every score records the model and the version of the standard that produced it. That is part of the record, not a footnote.

Models are retired by the companies that make them, so this is not a question we can avoid by picking one and keeping it. What we can do is refuse to mix judges without checking them against each other first.

Before any new model’s scores are shown, it is tested on sermons the current model has already scored, and we measure how far the two disagree — sermon by sermon, not on average.

It is only used alongside the old one if the two agree about as closely as one model agrees with itself. If they do not, the library is scored again with the new one rather than mixed — which is what we did in September 2026. Until a sermon is re-scored it keeps its earlier score, and its page says which model gave it.

So the library may hold scores from more than one model, and where it does, each score names the model that gave it. When a new model materially disagrees, that is a fact about our uncertainty and it should be visible rather than smoothed over.

11You cannot be corrected the way a pastor can.

This is the best objection on the page and we are not going to wriggle out of it.

It is true. A pastor is inside a relationship. He can be asked about it on Tuesday, he sits under elders, he knows the person he is preaching to, he can hear the objection in someone’s voice and stop. He can repent. Software can do none of those things, and no amount of engineering makes it able to.

Three honest responses, none of which dissolve the objection. First: we are not a substitute for that relationship, and if we ever become one for you, we have done you harm. The score is a starting point for a conversation with people who know you.

Second: what we can offer instead is transparency. A pastor’s reasoning is usually invisible; ours is printed underneath. You can see exactly what we based a finding on and tell us we are wrong.

Third: there is a person at the end of it. Corrections go to a published address and are read by a human being. We can be corrected — just not the way a pastor can, and not as quickly.

And the strongest form of this objection is one we would put more sharply than most critics do: we are not accountable to the preacher’s congregation, and they are the people most affected.

12Consistency is not a virtue if the standard is wrong.

Correct. A wrong standard applied consistently is worse, not better, than one applied erratically — it fails the same way every time and never has the human moment of doubt that stops it.

That is precisely why the standard is published rather than merely applied. Consistency is only worth anything downstream of a standard you can inspect and argue with. If we have got the standard wrong, we would rather find out from you than be reliably wrong at scale.

13You are scoring a machine’s guess at what was said.

True, and it is the largest uncontrolled variable in the whole system.

Every score rests on an automatic transcript, and automatic transcription mishears. A misheard sentence can become a finding. We would rather say that plainly than have it found out.

So we do not ask you to take the transcript on trust. Every finding shows you the words it was based on wherever the sermon has a line to quote, so if we have misheard, you will see it immediately — in the finding itself, without having to take our word for anything.

Every sermon page sends you to the church’s own recording. That is not decoration. It is how you audit us, and it is the same link we would point to if you told us we were wrong.

And the warning sits on every result rather than in the footer: the score assumes the text is what was actually preached, which is the one thing we cannot check for you.

14You withhold two of your four categories. What else are you not showing?

We show the two categories that describe a real divergence from Scripture, and we withhold two that do not.

The withheld ones are a passing slip rather than a teaching, and a reading the text will legitimately bear. We hold them and never publish them, deliberately. If we showed everything, a careful sermon would accumulate a list of minor observations that looked damning at a glance, and the length of the list would do rhetorical work the content did not deserve.

The honest tension is that you are being asked to trust that what we withhold is minor. The answer to that is the published standard, which describes all four kinds of thing we note — so you can see exactly what kind of thing is being withheld, even though you cannot see the individual items.

And the filtering happens on our server. Withheld material is never sent to your browser, because a filter that ships the data and hides it is not a filter.

15Three runs and you take the middle — is that not admitting it is unstable?

Yes, in the sense that any honest measurement admits its own error bars.

Scores up to 8.0 are run three times and the middle result is returned whole — not averaged, because averaging invents a result nobody produced. Higher scores are run once. The sermon page says how many times each was scored and, where it was more than once, how far apart the runs landed, so you can see for yourself how settled any particular score was.

We would be more worried about a system that produced an identical number every time and never told you how sure it was.

16So does 8.4 against 8.9 mean anything?

Barely, and we are the ones who measured it rather than waiting to be asked.

When we measured it in October 2026, across 294 sermons our current standard scored more than once, the highest and lowest results for the same sermon were typically about three tenths of a point apart, half a point on average, and in the worst case nearly four points apart. Where they are 0.8 or more apart, we show no score at all; the sermon is marked Not assessed. So half a point apart is within the noise. A whole point apart is probably real. And the score is genuinely useful across ranges — a 9 and a 5 are not the same sermon — while meaning very little at one decimal place.

We publish that because a tool whose limits you do not know is a tool you cannot use properly. And we would rather tell you the number than have you discover it.

17“Blind scoring” cannot be verified from outside.

Correct — you cannot check it from outside, and you should not simply take our word for it.

What we can tell you is that the identity of a sermon — church, preacher, title — is kept apart from the judgement, and the scoring engine is never given it (a recording can mention a name aloud, and the engine is told to pay it no attention). The two are brought together only after a score exists. That is not a policy someone has to remember to follow; it is how the thing is built.

And there is indirect evidence, from the one test that would expose a bias if we had one. In September 2026 we deliberately scored preaching from five traditions the library had never held, including Catholic and Black church preaching, and read what came back.

Two things in that result matter. Catholic homilies scored in the same band as the Reformed and Anglican preaching that dominates our sources — one of them above the Anglican average. And inside a single tradition the range was enormous: one Black church scored 9.2, 9.2 and 9.1 across three sermons while another scored 3.6, 4.0 and 5.5. An engine scoring the label could not tell those two apart. An engine reading the text can.

That is evidence rather than proof, and you are entitled to keep asking.

18Your library is biased by who you can reach.

Yes. Measurably — and we will tell you where, because you would find out anyway and it is better said by us.

What we hold is not a random sample of preaching. It is shaped by which churches make their sermons easy to reach, which is not the same thing as which churches preach well.

Two specifics we would rather say than have pointed out. Our coverage of prosperity and Word of Faith preaching was, until September 2026, almost non-existent — a handful of sermons from two preachers. Not by policy; by accident of what we had found. We went looking deliberately, because a service that quietly omitted the preaching people most often ask us about would be worthless.

And our coverage has been thinnest in Black church and Black-majority UK traditions. That is a consequence of how those churches most often publish their preaching, not of anything about the preaching itself — but a directory that let it stand would look like a judgement when it was really an accident, so we went after it directly, and where we cannot reach a church we ask the church.

A high average score in a region tells you about what we have reached, not about preaching there.

19This will be used to attack pastors, whatever you intend.

Some of it will be, and we cannot stop it. A number is portable in a way an argument is not.

What we can do is make the number expensive to misuse. The reasoning is free and always attached, so anyone circulating a bare score can be answered by anyone who follows the link. We score sermons, not people — one message at a time, dated, quoted. Every result carries a note saying plainly that preachers have bad weeks and unclear moments, that one sermon is a small window, and that the score is not an assessment of a preacher, a ministry or your church. And we do not rank preachers against each other.

But weigh this honestly: a tool that makes teaching easier to examine will sometimes be used by people who wanted a stick. We think the alternative — that public teaching should be harder to examine — is worse.

20A low score can damage a small ministry that cannot afford to answer.

A fair hit, and the asymmetry is real. A right of reply costs a large ministry nothing and a small one a great deal.

Our answer is partial and we will not oversell it. The right of reply is free, published, and goes to a person rather than a form. We publish the reasoning and the quotation, so a small church’s defence can be specific rather than a shouting match. And we do not add scores up into a verdict on a person or a church.

We do not think that fully answers the objection, and we are not going to write as though it does.

21You take money for judging work you did not create, from people who never agreed to be judged.

The money part has a clean answer. The consent part does not, and we will take them separately.

On money: nobody has ever had to pay to see a judgement about a preacher, and nobody ever will. The score and the reasoning are free to everyone, permanently, with no cap and no metering. A subscription buys convenience — saving without limit, following as many churches and preachers as you like, searching by length, the week in the pulpit, asking us to assess more from a church or preacher — never discernment. When someone asks us to assess more from a church or preacher, we choose which sermons, so a request can never single out one sermon. No church or preacher can pay to be assessed, to be assessed sooner, or to be removed. That was settled before there was anything to sell, precisely because a business that charged for verdicts would have a reason to produce more of them. You can test the claim in about ten seconds: sign out and read any score on this site.

On consent: this is the part with no comfortable answer. These preachers did not ask to be assessed, do not take part, and cannot opt out of being examined any more than an author can opt out of being criticised.

What we would say is that public teaching has always carried public accountability, and that this is heightened rather than lessened by scale. A person who teaches thousands of people — often as paid work, sometimes very well paid — has accepted responsibility for what they teach. Examining publicly published teaching, quoting it accurately, showing your reasoning and linking to the original is what criticism has always meant. We did not invent the scrutiny of preaching; the Bereans were commended for it.

22You say you score sermons, not pastors — but you list preachers by name.

Both are true, and the distinction is one we hold to in how the thing is built rather than merely assert.

Every score is of one sermon, on one date, on one passage. We do not produce an overall score for a preacher and we are not going to. We do not rank them, there is no league table, and the engine cannot see whose sermon it is when it scores.

But we do publish names, because the alternative is worse. An unattributed judgement about somebody’s teaching is impossible to check, impossible to answer, and impossible to correct. Naming the sermon is what makes the right of reply mean anything.

23Why should I trust you rather than my own elders?

You should not. Trust them. They know you, they are answerable to you, and they will still be there next year.

Use this the way you would use a concordance — as an instrument that helps you read carefully, not as an authority that tells you what to think. If something here troubles you about a preacher you love, the right next step is a conversation with someone who knows you both, not a screenshot.

24What happens when you are simply wrong about a specific sermon?

We are wrong sometimes. Here is what to do about it.

Write to the corrections address at the foot of this page. It is read by a person, not a queue.

Tell us the sermon and the finding. Because findings show the words they were based on wherever there is a line to quote, disputes are usually about something concrete — a mishearing, a quotation taken out of context, a passage misread. That makes them settleable rather than a matter of opinion against opinion.

While we look at it, we can take the assessment down at once. If we are wrong, it stays down until the sermon is assessed again.