Chapter 9: Algorithms, Penalties and Compliance

Quarterly compliance and penalty-defence review

1. Sweep the spam-prone surfaces. Internal search URLs and filter parameters should be noindex and should not be collecting inadvertent rankings. Search site:yourdomain.com plus the words you would never publish, and see what has been indexed under your name.
2. Retire the dead weight. Pages with no impressions, no clicks, no links and no purpose do not help you; they dilute the site. Merge, redirect or remove, on a schedule rather than in a panic after an update.
3. Harden the trust surfaces. Every article tied to a real author with a real profile, credentials, photograph and Person markup. Footer carrying a physical address, contact route, about page, editorial policy, privacy policy and, where relevant, returns policy. These are not decoration — they are the evidence a quality rater or a model uses to decide whether you are a real publisher.
4. Audit the link profile. New referring domains, anchor distribution, and sudden clusters from unrelated niches. Disavow only what is clearly manipulative and clearly yours to disavow — a disavow file built in panic usually removes links that were helping.
5. Protect your originals. Take two or three distinctive passages or images from your best pages and search for them. Scraped copies compete with you for the same query; a takedown request is often the fastest route back.
6. Log everything. A dated record of what shipped, when, and what the numbers did. When the next update lands, this log is the difference between analysis and superstition.

Interview questions and answer frameworks

Q1: The company published a lot of AI-written articles before you arrived. Under the scaled content abuse policy, what do you do?

Framework: Start with the distinction the platform itself makes: using AI is not the violation; publishing at scale without adding original value is. So the audit is about output, not tooling. Triage — pull every page published in that period, check impressions, clicks, conversions and whether the content contains anything that only your company could know: your data, your customers, your test results. Decide per page — pages with unique substance get kept, attributed to a real author, and illustrated with real artefacts; pages that are generic get merged into something better or removed. Change the process — a workflow where a practitioner defines the angle and the evidence, a model drafts structure and first pass, and a named human fact-checks, rewrites the generic passages and signs it. Say the part most candidates skip: the fix is expensive and slow, and pretending the old pages are fine is how teams get caught twice.

Q2: Traffic halved after a core update two months ago. What is your ninety-day recovery plan?

Framework: Four phases with explicit outcomes. Week one — post-mortem. Segment the loss: site-wide or a template, one market or all, brand or non-brand. Then study who took your positions and what they offer that you do not. Write the diagnosis down before touching anything. Weeks two to four — pruning. Remove or consolidate the pages that exist only to fill a calendar. On many sites the loss is not caused by the pages that dropped, but by the mass of thin pages around them. Weeks five to eight — rebuild the top of the portfolio. Take the thirty pages that historically carried most of the traffic and revenue and rewrite them properly: answer first, evidence added, comparison tables, real screenshots, named expertise, better structure. Weeks nine to twelve — external signals. Earn links and genuine community discussion for the rebuilt pages so the next reassessment has something new to weigh. Be explicit that nobody can promise a date for recovery; the plan is to deserve it, not to schedule it.

Q3: How do you distinguish an algorithm update from a manual action?

Framework: Manual actions are announced — they appear in Search Console with a reason and an example URL, and they require a reconsideration request. Algorithmic changes are silent: no notification, and the effect is distributed across many sites at once. So the diagnostic order is: check the manual actions report first, because it is free and definitive. Then check whether competitors moved too, and whether the timing matches a confirmed update window. Then check whether the drop is limited to specific query types or page templates, which often points at a change in how a surface is composed — AI answers, ads, a new result format — rather than a quality judgement about your site.

Q4: What is the compliance risk most teams underestimate?

Framework: Claim substantiation. Marketing copy that promises outcomes, health claims that need evidence, prices and specifications that go stale, and testimonials presented as typical results. It is a legal exposure that looks like a content problem, and it usually surfaces only when someone complains. The practical control is a claim register: what we assert, what supports it, who owns the source, and when it was last checked. On a reference site the same discipline applies to statistics — a number without a source and a date is a liability that gets quoted back at you.