SEO in the AI Era: A Practical Guide for Marketing Teams

Search has quietly split into two distinct channels. There is the traditional results page—ten blue links, featured snippets, and paid placements—and then there is the growing layer of AI-generated answers sitting above all of it. ChatGPT, Claude, Google’s AI Overviews, and Bing’s Copilot integration now answer questions directly, often without sending the user anywhere. For marketing teams, this creates a real tension: you can rank #1 on a keyword and still be invisible to the person who asked that question through an AI interface.

This guide is written for marketing teams who want to understand both channels and act on them without needing a developer for every change. Where you do need developer help, I’ve included exactly what to ask for.

The new search landscape

The clearest way to think about this split is to separate discovery engines from answer engines.

Discovery engines (classic Google and Bing) return a list of sources and let the user choose. Answer engines (ChatGPT, Claude, Google AI Overviews) synthesise an answer from multiple sources and may or may not cite where they got it.

Ranking well in discovery engines still matters—it drives direct traffic and it signals credibility to the AI systems that crawl the same web. But appearing in an AI answer requires something different: your content needs to be structured so that a language model can extract a clean, accurate answer from it, even if it only reads a single paragraph.

The underlying quality framework that connects both channels is Google’s E-E-A-T: Experience, Expertise, Authoritativeness, and Trustworthiness. ^[1] This isn’t a ranking signal you can game with meta tags; it is a description of the kind of content that earns links, gets cited, and gets pulled into AI answers. The practical implication: content written by someone with genuine subject-matter knowledge, backed by verifiable sources, consistently outperforms content that is optimised purely for keywords.

Content strategy: writing for humans and machines

The island test

Before publishing anything, apply what I call the island test: read a single paragraph in isolation and ask whether it still makes sense. AI models do not always read a full page before extracting an answer—they often pull a paragraph or a heading-plus-paragraph pair. If your paragraphs depend on context from three sections ago to be coherent, they will not extract cleanly.

Practically, this means:

• Define terms when you first use them, even if you’ve used them elsewhere on the site
• Avoid pronouns that reference earlier paragraphs (“as mentioned above”, “this approach”)
• Keep each paragraph focused on a single idea

Answer-first formatting

Google’s documentation on AI Overviews shows that the feature prefers content that leads with the answer. ^[2] The pattern is simple: state the answer in the first 40–60 words of a section, then provide the supporting detail. This is the opposite of how a lot of marketing copy is written, which tends to build context before arriving at the point.

Compare these two openings for a section on email frequency:

Before: “When thinking about how often to send marketing emails, there are many factors to consider, including your audience’s preferences, the type of content you’re sending, and your overall campaign goals...”

After: “Send marketing emails no more than twice a week for most B2B audiences. Higher frequency typically increases unsubscribe rates without improving conversion. The exceptions are transactional sequences and time-sensitive promotions.”

The second version passes the island test and leads with the answer. A model extracting content for an AI summary will almost always prefer it.

Citation density

AI models are trained to favour content that contains verifiable, specific claims—original statistics, expert quotes with attribution, and dated references. Generic statements (“many companies are adopting AI”) carry less weight than specific ones (“According to an October 2025 Gartner survey, 68 % of enterprise marketing teams had deployed at least one AI writing tool by Q3 2025” ^[3]).

If you have proprietary data from customer surveys, product usage, or industry reports, publish it. Even small datasets add citation value that generic content cannot match. When you cite external sources, link to the primary source rather than a secondary article that references it.

Technical essentials for non-technical teams

Robots.txt decisions: training vs. search

This is the most important technical decision marketing teams need to understand right now, and it is genuinely a strategic call rather than a purely technical one.

AI companies run two types of crawlers: training crawlers that collect content to train future models, and search crawlers that index content to include in AI-generated answers. You can block one without blocking the other.

If you want to protect proprietary content from being used in model training while still appearing in AI search results, your robots.txt should look like this:

User-agent: GPTBot
Disallow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ClaudeBot
Disallow: /

User-agent: Claude-SearchBot
Allow: /

GPTBot and ClaudeBot are OpenAI and Anthropic’s training crawlers. OAI-SearchBot and Claude-SearchBot are their search indexing crawlers. ^{[4] [5]} Blocking training while allowing search is a reasonable default for most marketing sites. Blocking everything means you opt out of AI search visibility entirely.

Ask your developer to check the current robots.txt file (it lives at yourdomain.com/robots.txt) and confirm whether any AI crawlers are currently blocked or allowed.

Structured data: what to ask your developer for

Structured data is code added to a page that tells search engines and AI systems exactly what kind of content they are looking at. You do not need to write it yourself, but you do need to know what to request.

The two formats most relevant to marketing content are:

FAQPage schema — for any page that contains questions and answers. Google explicitly recommends this format for content that may appear in AI Overviews. ^[2] Here is what the code looks like so you can recognise it when your developer shows it to you:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "How do I optimize content for AI search?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "To optimize for AI search, focus on 'answer-first' formatting, high factual density, and implementing structured data like FAQPage schema."
    }
  }]
}
</script>

Article schema — for blog posts, news articles, and long-form content. This tells search engines who wrote the content, when it was published, and when it was last updated. That last-updated date matters more than most teams realise—AI systems tend to prefer recent, maintained content over content that hasn’t been touched in years.

When briefing your developer, ask for: “FAQPage structured data on any page with a Q&A section, and Article structured data on all blog posts with author and date fields.”

Core Web Vitals: what marketers can control

Core Web Vitals are Google’s page-experience metrics. ^[1] Most of the heavy optimisation (server response times, JavaScript execution) requires developer involvement, but there are two things marketing teams typically control directly:

Image file sizes. The single most common cause of slow pages on marketing sites is uncompressed images. Before uploading any image, run it through a tool like Squoosh or TinyPNG. As a rule of thumb, no image on a marketing page should exceed 200 KB, and hero images should be under 400 KB.

Embedded third-party scripts. Every marketing tool you add—chat widgets, analytics tags, heat-map tools, A/B testing scripts—adds load time. Audit your tag manager quarterly and remove any tags that are no longer actively used.

Measuring success in the AI era

Keyword rankings still matter, but they are an incomplete picture. A page can drop from position 3 to position 7 in traditional search while simultaneously becoming a frequently cited source in AI answers—and the traffic and brand impact of the latter may exceed the former.

Metrics worth tracking

Brand citations in AI answers. Manually query ChatGPT, Claude, and Bing Copilot with questions your customers ask. Note whether your brand or content is cited. This is not yet fully automatable, but a monthly spot-check of your top ten target questions takes about 30 minutes and reveals a lot.

Bing Webmaster Tools AI Performance. Bing’s Webmaster Tools includes reporting on how your content performs in AI-generated responses within Bing Copilot. ^[6] If you haven’t verified your site there, it is worth doing—the data is distinct from what Google Search Console provides.

Crawl coverage. Check Google Search Console monthly for crawl errors and indexing issues. Pages that Google cannot crawl cannot appear in AI Overviews. ^[1]

Quarterly SEO health check

Run through this list every quarter; most items take under an hour total:

• Review robots.txt—confirm AI crawler rules are intentional
• Check Google Search Console for new crawl errors or manual actions
• Audit the five highest-traffic pages: do they pass the island test? Do they lead with the answer?
• Confirm structured data is present on FAQ and article pages (use Google’s Rich Results Test tool)
• Compress any images uploaded in the last 90 days that exceed 200 KB
• Remove unused third-party scripts from your tag manager
• Spot-check brand citations in ChatGPT, Claude, and Bing Copilot for your top ten target questions
• Update at least two high-value pages with fresh data or examples—the publication date signals recency to both search engines and AI systems

The fundamental shift in search is that quality and clarity have become more important, not less. AI systems are better than keyword-matching algorithms at detecting thin content, circular reasoning, and generic claims. The practices that help you appear in AI answers—specific claims, clear structure, verifiable sources, genuine expertise—are the same ones that have always produced durable search performance. The tactics have changed; the underlying principle hasn’t.