The short answer: same team, new assignments
Adding generative engine optimization to an SEO program does not require a new department. It redistributes work across three roles most teams already have. The SEO manager keeps the technical and ranking foundation and adds prompt-level measurement. The content lead keeps the editorial calendar and restructures briefs around the first question a buyer asks an AI engine. The PR owner moves from a supporting act to a formal part of the visibility program, because the citation data consistently shows earned coverage doing work that owned pages cannot. What genuinely changes is the measurement layer and the reporting rhythm: a small set of new KPIs alongside the old ones, and a cadence that respects how noisy AI answers are from run to run. Everything below is the operational detail of those three shifts.
Why you keep the SEO fundamentals
The case for keeping your SEO program intact is direct. Semrush's study of 5,000 queries and 150,000 citations found that Perplexity cited a domain from Google's top ten in 91% of cases and the exact URL in 82%; for Google AI Overviews the figures were 86% and 67%. Google's own AI-search guidance says the same foundational SEO practices apply to its AI features, and that no special AI text files or special schema are required. So the SEO manager's existing work, from crawlability to rank-worthy pages, remains the base of the program. The reason to add anything at all is that the overlap is not uniform: Ahrefs' test of 3,311 head terms found ChatGPT's overlap with Google's top results was far lower, 31.8% at the domain level and just 10% at the exact URL. Different engines reward the same fundamentals to very different degrees, which is why the team needs new instrumentation, not a new philosophy.
What the SEO manager does differently
The role keeps its core responsibilities: technical health, index coverage, and rankings on commercial terms. The additions are a second measurement layer and a recurring audit habit.
- Maintain a prompt inventory alongside the keyword list: the buyer questions you want to appear in, tracked per engine, because overlap with Google rankings varies widely between engines
- Adopt Google Search Console's generative-AI performance reports, announced in June 2026 for a subset of sites, as the free first-party baseline for AI Overviews and AI Mode by page, country, device, and date
- Run a monthly citation audit: which pages engines actually cite about the brand, and which weak or duplicated pages need consolidation
- Own entity consistency, making sure the brand name, descriptions, and structured data tell one coherent story across the site, docs, and third-party profiles
What the content lead does differently
The calendar does not get thrown out; the briefs get restructured. Profound's analysis of roughly 730,000 cited US-English conversations from late 2025 found that citation incidence fell from 12.6% at the first turn of a conversation to 4.5% by turn ten, so pages have to win the opening question a buyer asks, such as "what is," "best for," or "X vs Y," rather than counting on a long research session. Semrush's comparison also found commercial AI responses were roughly twice as long as informational ones, which argues for depth where money changes hands and brevity where it does not.
- Every brief opens with the direct answer, then evidence and caveats, in that order
- Commercial pages carry structured comparisons, current prices, trade-offs, and explicit best-for and not-for calls; informational pages open with a tight definition and stop
- Assign one owner per canonical fact, such as pricing or feature lists, so engines never find two conflicting versions on your own site
Why the PR owner gets a formal seat
This is the biggest structural change. Muck Rack's May 2026 analysis of more than 25 million links cited in ChatGPT, Claude, and Gemini across 17 industries found that earned media made up 84% of links while paid or advertorial content contributed 0.3%; journalism alone supplied 27%. Profound's 27-million-citation analysis adds that owned sources supplied only 4.3% of citations on category prompts. Read together: on the prompts where buyers discover a category, citations mostly go to coverage the brand earned, not pages it published or placements it bought. That makes PR a line function in the visibility program rather than a brand-awareness sidecar, with its output reported next to content and SEO metrics.
- Shift pitching toward trend-shaped stories: in Muck Rack's data, trend questions drew journalism at more than twice the rate of how-to questions
- Publish a reporter-ready methodology page alongside any data the company releases, so coverage links to something citable
- Track earned citations in AI answers in the same dashboard as content and SEO metrics, not in a separate comms report
KPIs: add a layer, do not replace the stack
Keep rankings, organic sessions, and conversions. They still describe most of the funnel, and the overlap studies show top-ranking pages still feed what AI engines cite. Add a small set of prompt-level metrics on top, and resist blending them into a single opaque "AI score" that hides which engine actually moved.
- Prompt coverage: the share of tracked prompts where the brand appears, broken out per engine
- Citation share: how often citations point at your own domains versus earned coverage about you
- Answer quality: whether engines describe the brand accurately and favorably, with errors routed to content or PR as fixes
- Lagging indicators: branded search and direct traffic lift, reviewed quarterly rather than weekly
A reporting cadence that fits the noise
AI answers are not stable: Ahrefs' citation research documents that identical prompts can produce different responses from run to run, so a single weekly snapshot will manufacture false alarms. A workable rhythm, and this is editorial guidance rather than a study finding, is a light weekly check for outages and major flips, a monthly review of coverage, citation share, and how engines describe the brand, and a quarterly refresh of the prompt set as buyer language shifts. As an illustrative example, a monthly review can be a twenty-minute agenda: the coverage trend per engine, the three worst answers read verbatim, and one owner with one fix assigned per answer. Leadership reporting works best monthly and trend-lined, with single-run swings explicitly excluded.
