Ask a candidate “how would you diagnose a 30% traffic drop” on Tuesday morning, and by Tuesday evening that question can be sitting in a public database with your company’s logo next to it.
Prepfully’s interview question bank is the clearest example of how this now works. Questions are submitted by candidates who have just finished a loop, tagged with the company that asked them, timestamped in hours rather than months, and sitting next to a tool that reviews a draft answer against what it calls company rubrics, trained on over a million submitted answers. Browse it for five minutes and you will see entries from Apple, Netflix, Disney, Siemens and a few hundred others, some posted eight hours earlier.
The roster behind it covers 138 roles across engineering, data and product management. There is no SEO role on there, which means your exact questions probably are not indexed yet. Treat that as a timing accident rather than protection. The mechanism works on any role with enough candidate volume to generate submissions, and the adjacent interviewers in your loop are already fully covered.
The thing this breaks is your scorecard, not your question list
Most hiring managers hear “the questions leaked” and reach for new questions. That misreads the problem.
What a candidate buys with a $300 mock session is not an answer. It is structure, pacing and the confidence to pause before answering. Those are the exact traits an unstructured interviewer mistakes for competence. So the real effect of a mature prep market is not that candidates know your questions. It is that the gap between your strongest and weakest candidate on delivery narrows, while the gap on actual judgment stays exactly where it was, and your scorecard can no longer tell the two apart.
Which means if your rubric weights communication equally with judgment, you are now substantially buying rehearsal. For an SEO role that is a worse trade than it sounds, because the job involves persuading engineers and finance people who do not care about search, and a candidate who presents beautifully for 45 minutes can still be the person who ships nothing in a year. Polish is necessary here. It is just not scarce any more.
Four swaps that survive being published
The fix is questions that are useless to a candidate who read them in advance, because answering them requires information only you have. Any of these can be posted on the internet tomorrow with no effect on their value.
| Instead of | Ask |
|---|---|
| ”How would you diagnose a traffic drop?” | Hand them a real annotated Search Console chart from your own property, with the legend removed, and ask what they would rule out first |
| ”How do you prioritise SEO work?” | Give them your actual backlog of eleven items and one engineer for two sprints, then ask what they would drop |
| ”What is your approach to content?” | Pull up three of your own underperforming pages and ask which one they would fix, keep or delete |
| ”Tell me about a time you influenced stakeholders" | "The CFO thinks organic is free. You have four minutes with her. Go.” |
The fourth one is the most predictive question I know of for an in-house role, and it cannot be prepared for in any useful way because the answer depends on knowing your business. It also surfaces the failure mode that matters, which is a candidate who starts explaining what a canonical tag is to a finance executive.
Strip the legend off that chart. Candidates who have rehearsed drop diagnosis will recite the framework. The ones who have actually done it start asking you questions, about release dates, about whether the drop is one page type or all of them, about whether impressions moved or only clicks. That instinct to ask before answering is the whole signal, and there is no way to coach it in isolation.
Pay for the work sample
For a senior role, nothing in a conversation beats three hours of real output. Send a scoped task, pay for it at a sensible rate, and make the deliverable small enough that it cannot eat a weekend.
Three hours at £75 to £150 an hour is £225 to £450 per finalist. Two or three finalists puts the whole exercise under a thousand pounds, against a fully-loaded cost of around £92,000 for the hire itself, which we broke down in fractional SEO versus a full-time hire. It is the cheapest insurance in the process and most companies still ask for it unpaid, which filters out the strong candidates first because they have options.
Good tasks are narrow. One page type, with the brief being what you would change and why, and a ceiling on the number of recommendations. Our audit template is the shape of a full engagement rather than a test; for a work sample, cut it to a single section and judge the reasoning rather than the coverage. Ask for the thing they would do first and what they would need from you to do it, which is the same thing the candidate side of this is rehearsing in interview prep for in-house roles.
One governance thing worth checking
Coaches on these marketplaces can stay anonymous. Prepfully lists each one by an ID rather than a name, says privacy is the coach’s choice, and lists referrals alongside mocks and advice sessions as something experts provide. Over 1,500 of them, from around 600 companies.
That is a legitimate side income and I have no problem with it. But it does sit in an odd spot next to two policies most companies already have, one on moonlighting and one on referral bonuses. Somebody in your organisation may be paid to coach candidates for roles at your organisation, and a referral bonus can in principle be paid on a candidate the referrer was also paid to prepare. Worth a sentence in a policy before it is worth a conversation with your legal team.
The practical step is smaller than that, though. Before you finalise a loop, spend twenty minutes reading the public bank for your own company name. Whatever your interviewers have been asking for the last two quarters is likely to be in there, and seeing your own questions quoted back by a stranger is the fastest way to work out which ones were never worth asking.
