SEO in the AI Search Era: From Keywords to Prompts, Brand Mentions and Pipeline

SEO is no longer measured by keyword rankings alone. A buyer may encounter a Google AI Overview, ask a full question in ChatGPT or Perplexity, verify the answer through third-party sources, visit the brand site and then contact sales. Teams therefore need to manage three connected assets: the questions buyers actually ask, verifiable brand mentions in credible sources, and qualified leads or pipeline influenced by search and AI discovery. Keywords, content, technical SEO and links still matter, but they now support a longer decision journey.

Ahrefs updated a study in 2026 using 300,000 keywords, comparing 150,000 queries that triggered AI Overviews with a control group. It found that when an AI Overview appeared, average CTR for the top-ranking page was about 58% lower than forecast. This is a correlation study, not a promise that every site loses 58% of its clicks. Query type, country, device, whether the feature appears and whether the page is cited can change the outcome. The practical response is to find queries in Search Console where impressions and position remain stable while CTR declines, then sample the live SERP before attributing the change to AI features.

Keyword data still reveals demand and search language, but AI users often ask complete questions containing context, constraints and desired outcomes. Collect the twenty questions buyers ask most often from sales calls, support tickets, site search, comments, query reports and discovery meetings. For each, record the audience, decision stage, constraints, required evidence and desired next action. Do not generate a thin page for every prompt variation. Consolidate questions with the same intent and create a separate URL only when the user task is materially different.

• Core question and natural-language variants
• Target market, language and buyer role
• Awareness, comparison, procurement or usage stage
• Direct answer and necessary qualifications
• Official sources, data, cases or expert review supporting the answer
• Relevant service, case, insight and contact paths
• Observation fields for Google ranking, AI citation, brand mention and enquiry

Backlinks still support discovery and help search systems understand relationships between pages, but link volume alone cannot explain whether an AI system will mention or recommend a brand. A stronger goal is to earn credible, relevant evidence through industry reporting, podcast interviews, expert commentary, original research, product reviews, professional communities and criteria-led lists. Unlinked mentions may form part of the wider evidence environment, but there is no public formula that guarantees one mention will produce an AI citation or recommendation. Record source quality, topical relevance, the exact claim, commercial disclosure and whether a reader can verify it.

• Does the source serve a real audience relevant to the topic?
• Why is the brand mentioned and what distinct value does it add?
• Are the facts, product names, company identity and service scope accurate?
• Are paid, gifted or commercial relationships disclosed?
• Is the page publicly accessible at a stable URL?
• Does the mention help a buyer decide rather than repeat a slogan?

A 3,000-word article is not automatically better than a 60-word answer, and the reverse is also true. Length should follow the complexity of the question. A useful structure answers directly in the first few sentences, then explains reasoning, conditions, steps, cases, risks and sources. Each section should remain understandable on its own without losing the article's central argument. Definitions, comparisons, procedures, FAQs and tables can improve clarity. Structured data must match visible content and cannot guarantee AI citation. Google also states there is no special AI schema, ideal AI-specific length or mandatory chunking format.

Search and AI systems can only process content they can access. Material product facts, use cases, methods, limitations, FAQs and authorised case summaries should not all sit behind forms. Yet 'publish everything' is not a responsible rule. Protect client privacy, unauthorised data, contractual terms, trade secrets and sensitive personal information. A two-layer model works well: public pages provide enough conclusions, methods and evidence for a buyer to evaluate the offer; personalised diagnosis, full datasets and internal implementation material can remain gated. Public content should create understanding and trust, not expose every asset.

Rankings and organic traffic remain useful leading indicators, but they do not prove commercial value on their own. Measurement should separate eligibility, retrieval, citation, mention, recommendation and conversion. GA4 can identify some referrals from ChatGPT, Perplexity and other platforms, but visits without a referrer may appear as Direct, while cross-device journeys, email, messaging apps and word of mouth create attribution gaps. Ask buyers how they first discovered the brand, then store first touch, last touch, landing page, question, requested service, qualification and opportunity stage in the CRM. Label buyer-reported discovery separately from technically confirmed referrals.

• Demand: coverage of high-intent keywords and prompts
• Visibility: rankings, impressions, AI-feature presence and cited pages
• Authority: quality of relevant media, community, review and brand mentions
• Visits: organic and AI referrals to priority pages
• Conversion: email, messaging, forms, bookings and qualified enquiries
• Business: opportunity value, wins, sales cycle and influenced pipeline
• Quality: source accuracy, content freshness and buyer feedback

• Week 1: collect the twenty most common buying questions and map them to existing pages
• Week 2: improve answers, evidence, FAQs and internal links on the three highest-intent pages
• Week 3: audit credible third-party mentions for factual inconsistency and authority gaps
• Week 4: configure GA4 referral grouping, key events and CRM source fields to establish a baseline
• Month-end: choose one topic to expand, one page to repair and one third-party evidence opportunity to pursue

SEO has not stopped working. What is outdated is using rankings and traffic to explain all value. Organisations still need crawlable websites, accurate pages, internal links and high-quality external recognition. They must also understand how buyers ask AI questions, where the brand is independently verified and whether discovery becomes a qualified enquiry or opportunity. WEPR connects question mapping, owned content, third-party evidence, sampled AI visibility and lead attribution in one workflow, without guaranteeing fixed citations, recommendations or rankings. The new scorecard should reconnect content with business decisions rather than create more vanity metrics.