AI Search Optimization

I start with a content audit focused on one question: does the page answer the user's core query directly in the first paragraph? For a client that runs a recipe site, I picked their most visited page for 'vegan chocolate cake'. The original led with a long story about childhood baking memories. No direct answer for at least 400 words. I rewrote the opening to say: 'This vegan chocolate cake takes 30 minutes to prepare and uses six ingredients you probably already have.' Then I added a FAQ schema with questions like 'How long does vegan chocolate cake last?' and 'Can I substitute eggs in this recipe?' I also sourced a factual claim: according to a 2023 survey by The Vegan Society, 72% of new vegans cite cake as the hardest dessert to find. That statistic got picked up by an AI Overview within two weeks, and the page saw a 3x increase in referral traffic from [AI chatbots](/chatgpt-seo/). The decision rule I now use is simple: if a page ranks top 10 for a question but does not offer an explicit answer in the first 100 words, that page should be restructured before anything else. Off-page authority is the other half of the equation. I run a mention audit using a tool like Semrush to count how often the brand appears in industry roundups, review sites and news. For that client, the brand had only 12 external mentions. I prioritised getting them into three recipe roundups. After that, [citations from AI systems](/ai-search-seo/) began appearing for related queries. My path is to fix the on-page answer gap first, then amplify brand mentions, and finally test prompts in ChatGPT, Perplexity and Google AI Overviews to confirm the page is being selected. I use a spreadsheet to log which prompts return which sources. Over three months, the client's citation count rose from zero to 14 across four AI surfaces.

One trap I see repeatedly is over-optimising a page for direct answers until it reads like a FAQ page without personality. I did this myself on a client site for 'how to clean a cast iron skillet'. The page became a list of bullet points. [AI systems cited it](/ai-seo/), but the human click-through rate dropped by 40% because the page no longer felt trustworthy. The lesson: write for people first; the AI will still extract the answer. Another watch-out is the assumption that schema markup guarantees a citation. I ran an experiment on 20 pages, half with schema, half without, all with the same direct answer. The schema pages were not cited more often. What mattered was the clarity of the heading and the proximity of the answer to the top of the page. [Off-page signals](/entity-seo/) can also backfire. A client with a negative review on Trustpilot saw their brand mentioned in an AI answer alongside that review. You cannot control how AI weighs sentiment. For niche topics with very few external mentions, AI systems often ignore your page entirely even if the on-page content is perfect. My rule of thumb: if your brand has fewer than 50 external mentions across the web, spend your first month on relationship building and guest posting before tweaking on-page structure. Finally, avoid relying on a single AI tool. I test across three surfaces because [each has different citation patterns](/llm-seo/). ChatGPT tends to cite authoritative news sources; Google AI Overviews often uses structured data; Perplexity favours pages with specific statistics. If you only optimise for one, you miss the others.

My biggest mistake was investing months into perfecting schema markup. I believed that if I marked up every page with FAQ, HowTo and Article schema, AI systems would naturally cite me. I was wrong. When I finally tested prompts in ChatGPT, I found that schema had almost no bearing on whether a page was cited. The citation came from the visible text that answered a question directly. I still add schema, but I now spend that time on [answer-first writing](/ai-copywriting/) and sourcing facts. A second error was ignoring off-page brand mentions. I had a client whose site was perfectly structured, but no AI chatbot would ever mention it. I checked the brand's mention count: less than ten external references. I had to go back and spend three months building citations through guest posts and industry roundups before the [AI surfaces](/ai-and-seo/) started picking them up. I now start every project with a brand mention audit. A third mistake was assuming AI search optimisation was a one-time fix. I restructured a page, saw citations for two months, and then they vanished. The AI models had updated their training data or changed their citation source. I now set a recurring monthly task to test prompts and re-audit the pages. The lesson: treat this as ongoing maintenance, not a project with an end date.