Who Owns AI Visibility: A RACI for Marketing and Dev

AI visibility touches four stages, and each sits with a different team. What a model knows without browsing (parametric knowledge) is frozen at a training cutoff, and nobody controls it this quarter. Whether a crawler can fetch your page (access) is controlled by your site and your CDN. Whether your page gets pulled into an answer (retrieval) and then used or cited (generation) is controlled mostly by the engine.

That means the controllable work is split. Access is a developer problem. The facts on the page, and the other sites that describe you, are a marketing problem. Deciding what to measure and what to expect is a leadership problem.

The evidence also tells you where effort pays. A July 2026 critical survey of 45 GEO studies found no technique with a stable, cross-platform causal effect on discoverability. The levers that hold up are access levers, and they work by removing blocks. So a good RACI gives developers clear ownership of the blocks, and stops marketing from buying tactics the evidence does not support. For the wider picture, see GEO explained honestly.

R is responsible (does the work). A is accountable (signs off). C is consulted. I is informed.

Two rules keep this table honest. First, every row has one A. Second, the C entries are real meetings or tickets, not courtesy copies. Marketing decides which facts matter. Developers decide how those facts reach crawlers. If your team is new to this split, AI SEO training for teams covers the shared vocabulary.

Developers: access and rendering

• Rendering. Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but do not execute them. Content that only appears after JavaScript runs may be invisible to those crawlers. Googlebot renders, so this gap mostly affects non-Google engines.
• Per-bot robots.txt. OpenAI's crawler page says GPTBot governs training use, OAI-SearchBot governs ChatGPT search, and each setting is independent. The same split exists elsewhere, for example Claude-SearchBot for Claude's search.
• Network reachability. OpenAI's help page says eligibility for ChatGPT search means allowing OAI-SearchBot and confirming your host or CDN allows its published IP addresses. A firewall can block a bot that robots.txt allows.
• Snippet controls. Google documents noindex, nosnippet, max-snippet and data-nosnippet as the on-page controls that keep content out of AI features.

Marketing: facts and third-party listings

• Maintain one list of canonical facts: prices, plan names, product names, locations, founding year.
• Make sure each fact is stated plainly on a page that can be fetched. A specific sentence gives an engine something it can quote.
• Remove contradictions between the pricing page, about page and footer.
• Keep third-party listings current. When engines cite other sites instead of yours, being named accurately on those sites is associated with appearing in the answer. See your own site vs third-party mentions.

Leadership: reporting expectations

• Decide which engines get measured, and measure each one separately.
• Set a cadence and stick to it.
• Refuse targets for numbers nobody can verify. As of October 2026, Google's AI reporting in Search Console shows impressions only, with no position data, so an "AI rank" has no ground truth.

Each of these is a failure between two owners. None is visible from inside one team.

1. Marketing moves a fact into JavaScript. A new pricing widget loads prices client-side. The page looks fine in a browser. Crawlers that do not run JavaScript see an empty table.
2. A developer adds a blanket AI block. The goal was to stop training use, so the rule blocks every AI user agent. Blocking OAI-SearchBot also removes ChatGPT search eligibility, which nobody asked for.
3. A CDN default changes without anyone editing anything. From 15 September 2026, Cloudflare blocks Training and Agent crawler categories by default on ad-displaying pages for new domains, new sites on existing accounts and existing free-tier customers. Security owns the setting. Marketing feels the effect.
4. A template inherits a suppression tag. A nosnippet or data-nosnippet tag added to hide a cookie banner ends up wrapping real content. The same thing happens with noindex copied from a test environment, which is also how staging sites end up in AI answers when the reverse mistake is made.
5. Leadership asks for a number that does not exist. A request for "our AI ranking" pushes the team toward vendors selling unsupported tactics, or toward content automation shortcuts. In Otterly's 2026 experiment, Google deindexed both mass-produced AI sites, with no manual action reported.
6. Facts change on the site but not off it. Marketing updates pricing on the website. Directory and review listings still show last year's plans, and engines that cite those sites repeat the old price.

Block 90 minutes with one person from each team.

1. Write the canonical facts list (marketing). Ten to twenty facts is enough to start.
2. Fetch your five most important pages as raw HTML with JavaScript off, and confirm each listed fact appears (developers).
3. Read robots.txt line by line for each AI bot: GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot, Google-Extended. Write down what each rule intends (developers, marketing consulted).
4. Request a page as each AI crawler and record the status code. A 403 or a challenge page means the firewall disagrees with robots.txt. The AI Crawler Checker runs this request per crawler.
5. Search the templates for noindex, nosnippet, max-snippet and data-nosnippet, and confirm each one is intentional (developers).
6. List the third-party pages that describe you and check them against the facts list (marketing).
7. Ask three buyer questions on each engine you care about and note which facts come back wrong (marketing). The free checker on this site gives a snapshot for up to three questions.
8. Agree the reporting cadence and success criteria, then fill in the RACI table with names (leadership).

If you only have one afternoon, the one-afternoon AI visibility check compresses steps 2 to 7.

• Every deploy that touches templates, rendering or CDN config pings marketing.
• Every new or changed fact (price, plan, product name) pings developers with the page it lives on.
• robots.txt changes name the specific bot and the intended consequence in the ticket.
• Security and CDN owners review bot-management defaults after any vendor announcement.
• Key facts are confirmed present in raw HTML after each pricing or product launch.
• Suppression tags are reviewed in any new template before release.
• Third-party listings are updated in the same sprint as the website.
• Google Search Console and Bing Webmaster Tools are checked separately, because one being healthy says nothing about the other.
• Reports show each engine on its own line, with the date and the questions asked.

Good results

• Every AI crawler you intend to allow gets a 200 response on your key pages.
• Raw HTML contains the same facts as the rendered page.
• Wrong facts in AI answers trace back to a specific source, and someone owns fixing it.
• Reports compare engines separately and over the same questions each time.

Bad results

• robots.txt allows a bot that the firewall blocks.
• Nobody can say who approved the last robots.txt change.
• The team reports one blended score and cannot explain why it moved.
• Goals are framed as citations or rankings promised by a date. Nobody controls what the models say, so these goals reward guessing.

Cadence

• Access checks: monthly, and after any deploy touching templates, rendering, robots.txt or CDN rules.
• Facts and listings: quarterly, and after every pricing or product change.
• Leadership review: quarterly. Re-read this RACI, confirm names, and drop any metric that cannot be verified.
• Vendor changes: whenever a crawler operator or CDN announces a new default. Platform behavior changes, so treat every finding as dated.

Do we need a dedicated GEO hire to own this? Usually not. The work maps onto roles you already have. Developers handle crawler access and rendering, marketing handles facts and listings, and leadership sets reporting rules. What you need is a named owner for each row of the RACI and a standing review. A dedicated hire makes sense only when the volume of facts, products or markets is too large for marketing to keep consistent.

Should the SEO team own AI visibility? The SEO team is a natural coordinator, because index eligibility and snippet controls overlap with classic search work. Google says pages must be indexed and eligible for a snippet to appear in AI Overviews or AI Mode. But SEO rarely controls firewall rules or CDN defaults. Give SEO the coordinating role and keep developers accountable for access.

Should marketing ask developers to add schema or an llms.txt file? Not as an AI visibility fix. Google says structured data isn't required for generative AI search, and Ahrefs' matched test found AI Overview citations fell 4.6% after adding schema. Google Search ignores llms.txt, and no engine documents reading it. Existing markup does no harm, so there is no need to remove it. Spend developer time on rendering and access instead.

How does this RACI work when an agency is involved? Put the agency in the R column for the work it actually does, such as audits or listings updates, and keep accountability in-house. Agencies rarely have access to your CDN, firewall or Search Console settings, so those rows stay with your developers. Make the agency report engine by engine, with dates and the exact questions asked.