Six SEO Interview Questions ChatGPT Can't Answer for Your Candidate

A hiring-managers thread on Reddit asked a decent question this week: if you were actually interviewing an SEO person, what would you ask them?

Most of the replies were fine. A few were sharp. The interesting part wasn’t the questions people suggested, it was the reason the old questions stopped working.

Ask a candidate what a meta description is. Ask what a 301 redirect does. You’ll get a correct answer almost every time now, because ChatGPT gives a correct answer to both in about four seconds, and anyone prepping for an interview has already asked it the same things you’re about to ask them. Definitional questions test whether someone read the same primer everyone else read. They don’t test whether the person sitting across from you can actually do the job.

What still works is a scenario with no clean textbook answer. Here are six.

”Traffic dropped 40% this month, but conversions went up. Is that a problem?”

The weak answer panics about the traffic number. The strong answer asks a question back before answering at all: which pages, which kind of traffic, branded or non-branded, what intent were those visitors searching with.

A 40% drop in low-intent informational traffic that never converted anyway, paired with a conversion increase, might mean the site lost visibility on queries that were never buying anyway while gaining it on queries that actually convert. That’s not a problem. That’s the site getting healthier. A candidate who reflexively treats “traffic down” as bad news, full stop, is pattern-matching on a headline number instead of thinking about what the number is made of.

”This page has zero backlinks and has ranked #1 for three years. Why?”

There’s no single correct answer here. That’s the point.

Someone reciting a checklist will list ranking factors in order and stop. Someone who actually understands search will start generating real hypotheses: maybe the competition for that query is genuinely thin. Maybe the page answers the query so completely that nothing else needs to exist. Maybe it’s benefiting from domain-wide authority the site earned elsewhere, and this specific page never needed its own links.

The answer that reveals the most is the one that names two or three competing explanations and says how you’d actually check which one is true, rather than the one that picks a single cause with total confidence.

”You can fix 50 minor technical errors or rewrite 5 strategic pages this month. Which do you pick?”

This is the question that separates someone who runs audits from someone who runs a business.

Fifty minor technical errors sound urgent because an audit tool flagged fifty things. Most of them move nothing. Five strategic pages, rewritten well, on queries that actually drive revenue, usually move more than fifty broken alt tags combined. A candidate who picks the fifty errors because “you should always clear technical debt first” is optimizing for a clean report, not for the client’s revenue.

”A client wants to show up when people ask ChatGPT or Perplexity about their industry, not just rank in Google. What’s the first thing you check?”

This is the newest question on the list, and it’s the one most candidates will fumble, which is exactly why it’s useful.

A candidate stuck in classic SEO thinking will start talking about keywords and backlinks. Someone who’s kept up will start talking about whether the site’s claims are specific and verifiable enough to extract, whether the technical documentation says something a competitor’s homepage doesn’t already say in different words. That’s the actual mechanism behind information gain optimization: AI systems cite the source that adds something they haven’t already seen, not the source that ranked best under the old rules.

”A competitor just added FAQ schema to every page. Should the client copy them?”

This one has a trap built into it.

The dated answer is yes, because FAQ schema used to earn a blue dropdown box in Google’s results, and more visible real estate was worth chasing. Google retired FAQ rich results in May 2026. That specific reason to add FAQ schema is gone.

The current answer is more precise: FAQ schema can still be worth adding, not for the rich result, but because structured data helps AI Mode verify a claim during answer generation, and only when the schema states something that’s also visible in the page’s actual content. A candidate who says “yes, obviously, FAQ schema is good practice” without knowing that distinction is running on outdated information and doesn’t know it yet.

”Here’s a page ranking #1 in Google that never gets cited in AI Overviews for the same query. What’s the gap?”

Ranking well and getting cited well are increasingly two different skills, and this question tests whether a candidate has noticed.

A page can rank first because of backlinks, domain age, and years of accumulated trust signals, and still lose the citation to a page that states a specific fact more plainly, in a format an AI system can lift directly. The gap is usually extractability: the ranking page might bury the actual answer in the fourth paragraph after two hundred words of preamble, while the cited page states it in the first sentence under a heading that matches the question almost exactly.

Telling a real answer from a rehearsed one

None of these six questions have a single correct answer memorized in advance, which is deliberate. What you’re actually listening for during the answer matters more than which conclusion someone lands on.

Red flags: an answer full of the right vocabulary and no specifics. Someone who can’t produce a single real number from actual past work when you ask a follow-up. Someone who answers every scenario with “it depends” and then never says what it depends on.

Green flags run the other way. A candidate who asks what data is available before committing to a diagnosis. Someone who volunteers a specific, real example without being prompted for one. And, counterintuitively, someone who says “I’m not sure yet, here’s how I’d find out” on the AI-search questions rather than bluffing a confident answer, because right now almost nobody has this fully figured out, and false confidence there is its own red flag.

What checkable proof of experience actually looks like

SEO has an evidence problem that most other skilled work doesn’t. A developer can point to shipped code. A designer can point to a portfolio anyone can look at and judge. SEO results live inside a client’s private analytics account, so most of what gets presented as proof is a self-written case study page, which is exactly as verifiable as it sounds: not very.

The alternative is a track record that lives somewhere neither party controls. Shashank Dubey’s Upwork profile is a decent example of what that looks like in practice: Top Rated Plus status, 100% job success across 17 completed engagements, more than 15,000 logged hours, working SEO and Google Ads for law firms and eCommerce and WordPress businesses. None of those numbers can be edited after the fact by the person claiming them. The hours are logged by the platform. The success rate comes from client ratings, not self-assessment.

That’s not a claim that a high Upwork number automatically means someone is good. It’s a claim that a number neither side can quietly inflate is worth more in a hiring conversation than a case study PDF with the client’s name redacted “for confidentiality.”

The actual point

None of this is about finding trick questions to catch people out. It’s about noticing that the entire category of question that used to feel rigorous, the kind with a single correct definition, stopped being rigorous the moment a free chatbot could answer it as well as any candidate.

What’s left is judgment. Ask about the traffic drop that isn’t actually bad news. Ask about the page that ranks with nothing behind it. Ask about the fifty small errors against the five big ones. Those questions still work, because the interesting part was never the definition. It was always what someone does with an ambiguous situation and incomplete data, and no model answers that one for them.