AI and SEO

I start with keyword research. I feed a seed keyword — say "vegan protein powder" — into an AI tool and ask for 200 related questions, long-tail variants and co-occurring terms. That returns a list I can cross-check with Google Search Console. Last month I found 40 queries that had impressions but zero clicks; I built content around those and three of them hit the top five within two weeks.

The second step is content outlines, not drafts. I ask the AI to produce a structure based on the questions it surfaced, including H2s, H3s and a brief for each section. I then take that skeleton and rewrite the angle. For example, the AI suggested "Benefits of vegan protein powder" — I changed it to "Why I switched to pea protein after 10 years of whey." That personal twist made the piece perform three times better on average session duration.

For on-page tasks I use AI for meta titles, meta descriptions and alt text. But I have a decision rule: if the output can be verified against a fact or a brand guideline within 30 seconds, I let the AI keep the first draft. If it requires tone judgement or competitive nuance, I write from scratch. That rule alone cut my editing time by half.

I also run technical audits with AI tools for crawls and broken link detection. The catch is that I always spot-check the output. AI crawl tools often miss JavaScript-rendered links or misclassify redirect chains. I once let an AI tool suggest schema markup for an FAQ page — it generated a nested QAPage that violated Google's structured data guidelines. I caught it only because I reviewed the test in the Rich Results tool.

For AI search surfaces — Google's SGE, Bing Chat — I structure content to answer the query directly in the first paragraph. I strip out unnecessary JavaScript and avoid "read more" truncation. The same fundamentals that work for featured snippets work here: use bullet lists, tables and clear declarative sentences. I still invest in [digital PR](/aiprm-for-chatgpt/) and topical authority because AI search models rely on the same crawlable web content that classic search does.

The biggest mistake I see is publishing AI-generated content without review. I have watched sites lose rankings because the AI produced plausible-sounding nonsense — a blog claiming that "the Great Wall of China is visible from space," which is false. Even when facts are correct, the writing often reads like a committee agreed on blandness. Your unique perspective is what makes content worth linking to. If you take that out, you are competing on keywords alone, and you will lose.

Another watch-out: using AI for everything. It is tempting to automate competitor analysis — feed in five competitor URLs, get back a report of their content gaps. But that report is only as good as the AI's understanding of your market's nuance. It might flag a gap that does not exist because it mistook a brand's positioning for a missing topic. I use AI for competitor analysis only as a starting point; I still manually read the competitor's top pages to understand their tone and audience.

Edge case: AI and schema markup. I have seen practitioners generate JSON-LD with AI and deploy it without testing. The AI often invents properties or uses outdated vocabulary. For instance, it might add "mainEntityOfPage" in a way that confuses the crawler. Always validate generated schema against Google's Rich Results test. If the tool you use does not support live validation, do not use it.

Also, do not let AI manage your internal linking strategy autonomously. I tested a tool that suggested links based on keyword similarity. It linked a page about "SEO automation" to a page about "AI copywriting" because both shared the word "AI". That made sense semantically but the user journey was disjointed. I now review every AI-suggested internal link and ask: does this help the reader move logically? If the answer is no, I skip it.

Finally, be wary of AI search optimisation advice that claims you need a separate strategy. The [best AI SEO tools](/best-ai-seo-tools/) still rely on core web vitals, clear headings and topical depth. Do not let a new buzzword distract you from the fundamentals.

I started by asking AI to write full blog posts. The output was technically correct but flat — no personality, no unique insight. I spent weeks editing mediocre drafts that I could have written better from scratch in half the time. Now I use AI only for outlines, meta tags and repetitive tasks. The real value is in the editing and strategy I bring. I also changed my mind about AI for [entity SEO](/entity-seo/). I assumed that because AI models understand entities, I could feed it a list of entities and get back a perfect content plan. But the AI often over-weighted generic entities (e.g., "protein" for a vegan supplement site) and missed niche entities (e.g., "pea isolate" vs. "soy isolate"). Entity research still requires a human who knows the domain.

I also got wrong the role of AI in search visibility. I thought that as AI-powered search grew, traditional SEO tactics like backlinks and domain authority would matter less. I tested this by comparing two pages: one with strong topical authority and few backlinks, another with many backlinks but thin content. The authoritative page outranked the backlink-heavy one for AI-generated search answers. That convinced me that [LLM SEO](/llm-seo/) is not a separate discipline — it rewards the same depth and clarity that good SEO always has.

Another error: I used AI to rewrite old content without checking whether the original still satisfied search intent. The AI turned a practical "how to" article into a fluffy overview because it summarised the topic at a high level. Always check intent before asking AI to refresh content. If the intent is commercial, keep the specifics and the comparison tables; don't let AI soften them.

Finally, I underestimated how much AI tools can vary in quality. I tried five different AI content generators and found that each introduced its own bias — one overused industry jargon, another avoided any strong claims. You need to calibrate the tool to your brand voice, and that takes time. Do not assume one model fits all.