Interview Questions for SEO and GEO Roles
The questions below repeat across interviews because they separate people who have run a site from people who have read about running one. Each answer is a framework, not a script — interviewers are listening for how you structure the diagnosis, and for whether you name a mechanism instead of a tool.
Technical questions
Q: What is the difference between the title tag and the H1?
The title tag is the label in the search result and the browser tab; it competes for the click. The H1 is the heading at the top of the page and competes for the reader’s attention once they arrive. They should be related but not identical — identical wastes a chance to target a variation, and unrelated confuses both readers and engines about what the page covers.
A worked pair: the title could read “10 Best Coffee Makers 2026 (Tested and Reviewed) | Brand”, while the H1 reads “The 10 Best Coffee Makers of 2026, Tested by Our Baristas”. Same promise, different job.
Q: When do you use a 301 redirect, and when do you use a canonical?
A 301 is an instruction: the old URL is gone, all signals move. A canonical is a suggestion: this URL is accessible, but prefer that one.
Use a 301 when a page is permanently retired and a real replacement exists. Use a canonical when several URLs must stay live — sort orders, filter parameters, tracking variants — and you want signals concentrated on one. The rule that catches candidates: never combine noindex with a canonical on the same page, because the two signals contradict each other and the outcome depends on which engine reads them first.
Q: What are the most common hreflang mistakes?
Five, in roughly the order they appear in the wild. One-way annotations, where A points to B and B does not point back — the cluster is ignored. Confusing language with region: the region code for the United Kingdom is gb, not uk, and the language code for Japanese is ja, not jp. Every locale’s canonical pointing at the English master instead of itself, which suppresses the whole set. Omitting x-default, so there is no designated fallback. And IP-based automatic redirection, which means the crawler can only ever see one version of the site.
Q: How would you make a React or Vue site indexable?
The problem is that a client-rendered single-page app serves a nearly empty HTML document, so anything that does not execute JavaScript sees nothing. In order of preference: server-side rendering with Next or Nuxt; static generation where content does not change per request; prerendering for crawlers if the stack cannot change; and dynamic rendering as a last resort, because maintaining two rendering paths doubles the ways the page can break.
Verify with URL Inspection, reading the rendered HTML and the rendered screenshot rather than the source. If the content is in the screenshot and not in the HTML, you have found the bug.
Q: An ecommerce site has half a million faceted URLs. What do you do?
Block by default, then allow a whitelist. Facet combinations multiply into crawl traps, and almost none of them correspond to how anyone searches. Four steps: suppress multi-facet combinations in robots or meta robots; research which category-plus-one-attribute combinations actually have demand; give those a clean static path with its own title, H1, intro copy, and self-referencing canonical, and put them in the sitemap; then monitor the index to confirm that only the whitelist is being crawled and stored.
The mistake to avoid is the reverse order — opening everything first and trying to close it later, after the crawl budget and the index have already been spent.
Strategy questions
Q: A high-traffic page converts far below the site average. How do you diagnose it?
Start with intent before touching the page. Which stage of the decision is this query from? If the page is absorbing top-of-funnel research traffic, a low conversion rate is the correct outcome, and the fix is a different page, not a better button.
Then work downward. Watch recordings and scroll depth to find where attention dies — usually above the fold, or at the exact point where the page switches from answering to selling. Read the first screen as a stranger: is the value proposition and the next step visible within three seconds? Audit the call to action: does it name a concrete gain, or does it say “learn more”? Finally, test one change at a time on the weakest step, with enough traffic for the result to mean something.
Q: How do you decide between refreshing old content and writing new content?
Rank candidates on two axes: current position and distance from being competitive. Pages at positions 4–15 with real impressions and no fundamental content gap are the cheapest wins available — a data refresh, a new section, a FAQ block, a rewritten title often moves them several places. Pages that used to rank and have declined over six months are next, because the decline is evidence of a specific problem you can find. Pages at 16–30 need substantial upgrading, and pages past 50 with no impressions are a rewrite-or-delete decision based on whether the topic supports the business at all.
Q: Your boss asks you to take domain rating from 20 to 40 next month.
Manage the expectation rather than accepting it, and explain why. Domain rating is a third-party proxy for link authority; the search engine does not use it. Movement of that size typically takes six to twelve months of consistent acquisition, and any vendor promising it in a month is selling links that will not survive.
Then redirect to what the number was standing in for: how many referring domains from relevant sites were added this month, how core keyword positions moved, and what organic pipeline resulted. Authority is an outcome. Treating it as a target is how teams end up buying links.
Q: A core update just landed and organic traffic is down 25%. What are your first moves?
Confirm causality before diagnosing. Does the decline date align with the rollout, or with a deploy, a tracking change, or a seasonal shift? Exclude those first — the most common false alarm in this job is a tagging bug that looks like an algorithm penalty.
Then scope it: compare fourteen days before and after in Search Console, and identify which directories, page types, and query groups absorbed the loss. Compare winners and losers on the same queries — the pages that took your positions are the specification for what the evaluator preferred. Classify the cause as content quality, link profile, technical regression, or intent drift, because each has a different remedy. Then execute over ninety days and wait for the next update to measure recovery, since interim noise will not tell you whether the fix worked.
GEO questions
Q: What is GEO, and how does it relate to SEO?
GEO is the practice of making a brand more likely to be retrieved, cited, and recommended by generative systems — ChatGPT, Perplexity, Copilot, Google’s AI Overviews.
It sits on top of SEO rather than replacing it. Retrieval-augmented systems still begin by searching an index, so a page that cannot rank conventionally is not a candidate for citation. The difference is the unit of optimisation: SEO optimises a position, GEO optimises an appearance inside a generated answer, which may cite three sources, one, or none.
That shift changes what matters: factual density, structure that survives being cut into passages, consensus across third-party sources the brand does not control, and clear entity signals so the model knows what the brand is.
Q: A brand ranks top three on Google but is never mentioned by ChatGPT. How do you diagnose that?
Work down four layers. Retrieval: is the site in the index the assistant actually uses, and is it blocked by robots rules aimed at AI crawlers — often added by a security plugin nobody remembers installing. Chunking: does the markup give the extractor clean boundaries, or is the content trapped behind interaction and split mid-sentence? Synthesis: does the page contain quotable specifics — numbers, comparisons, a direct answer in the first sentence under a question heading — or only narrative? Consensus: for recommendation questions, assistants prefer third-party evidence, so check whether the brand appears in the review sites, communities, and comparison articles the model is likely to sample.
Most “invisible in AI” cases are the fourth layer, and no amount of on-site editing fixes it.
Q: How do you measure GEO?
Four layers, because each answers a different question. Visibility: how often the brand is mentioned across a fixed prompt set, and where it ranks within the answer. Citation: how often your URLs are linked, and whether the facts attributed to you are correct. Traffic: sessions arriving from assistant domains, judged by revenue per session rather than volume. Business: branded search lift, self-reported attribution at checkout, and pipeline traced back to the programme.
Fix the prompt set in advance and keep it stable, or the trend is meaningless. And publish the methodology alongside the number, because this measurement is new enough that any figure without a method invites disbelief.
Communication and commercial questions
Q: How do you explain to a non-technical CEO why SEO deserves budget that would otherwise go to paid search?
Use a comparison that runs over two years, not one month.
“At twenty thousand a month in paid search we buy clicks, and the month we stop, they stop. At the same spend on content, links, and technical work, months one to three look like nothing. By month six we might be buying the same traffic at a similar effective cost, which is the point where most companies quit and conclude it didn’t work. By month twelve the content is still producing while the ad budget would have to be spent again, and the gap widens from there because the asset keeps working with no additional spend.
The honest version: SEO is slower and it is not free. What you are buying is an asset with a long tail instead of a rental with a short one.”
Q: Product and engineering deprioritise every SEO request. How do you get them shipped?
Three approaches, used together. Translate the request into their metrics: not “improve LCP” but “LCP is 4.2 seconds, which puts us in the poor band, and the poor band correlates with a measurable share of the traffic we are losing”. Reduce their cost of saying yes: a ticket with the exact change, the test, and the acceptance criterion takes an engineer twenty minutes; a document explaining why SEO matters takes an afternoon. And build reciprocity: bring them something they want — a query pattern that reveals an unmet product need, or evidence that a competitor’s feature is being searched for by name.
Influence here is accumulated, not asserted. The person who arrives with a completed ticket gets help next time.
Q: You inherit a link profile built entirely from purchased links. What do you do?
Assess before acting. Quantify the proportion of obviously manipulative links and check for any manual action in Search Console. If the share is small and there is no manual action, the practical answer is usually to leave it and build better links on top — aggressive removal campaigns often cause more volatility than the problem.
If the share is large or a manual action exists, attempt removal at the source, disavow what cannot be removed, and document every step in case you need to explain the history. In parallel, begin acquiring links that would be defensible in public: original research, tools, expert commentary. And write down a link policy, so the next person inherits a rule instead of a cleanup.