Communications is well placed to lead AI visibility, because the way in which AI systems describe a firm is a reputational issue. The role needs close collaboration with SEO and digital teams, covering both third party sources and the firm’s own content. In a survey of 971 PR professionals published in July 2026, 29 per cent said nobody internally owned generative engine optimisation and 39 per cent were not measuring it. Here is the job description to fill that gap.
How wide is the AI visibility accountability gap?
The gap is wide, and it is measurable. The Muck Rack 2026 State of PR was published on 9 July 2026, drawn from 971 qualified responses collected between 14 May and 12 June 2026. It found that 73 per cent of PR professionals rated AI search visibility as at least somewhat important in the 2026 survey. That same 9 July 2026 study found that 29 per cent said nobody internally owned generative engine optimisation and that 39 per cent were not measuring it at all.
An important function without a clear owner falls between desks. The answers also move. Ask the same question of several AI assistants and the firms they name differ, and the same assistant can change its answer from one week to the next, which is a strong argument for a standing owner rather than one-off audits. Our AI visibility audit checklist sets out how to run those checks.
Why does this sit in communications rather than in SEO?
AI visibility sits in communications because the inputs that shape it are earned sources. AI answers can draw on news coverage, reference sites, forums and firms’ own websites - all areas in which communications play a role in shaping reputation.
SEO owns the pages you publish and the technical health of your own domain. That work helps search-based AI services find and use the firm’s content. Measurement in share of voice has changed: traditional metrics often assumed one ranked list, whereas AI visibility requires a different methodology tested against multiple models over time. Where the main objective is reputational accuracy, communications needs to own the problem.
What are the six responsibilities of the role?
The role carries six responsibilities. A written remit makes responsibilities and escalation routes clearer.
- Monitor agreed prompts across the AI services relevant to the firm’s audiences, recording the service, available version information, search mode and testing date.
- Investigate material inaccuracies, correct owned information and request evidence-based corrections from relevant third-party publishers.
- Use observed citation patterns to inform media research, alongside audience relevance and editorial credibility.
- Own the four metrics below and report them on a fixed cadence.
- Coordinate with SEO, digital and legal so owned pages and structured data are in alignment.
- Keep the tooling current and the method documented, with a quarterly review of the vendor landscape.
Who should the role report to, and who else needs to be in the loop?
For a role focused on reputation, a reporting line into communications is a practical starting point. The owner needs regular contact with four other functions, each of which influences the effectiveness of the role.
- SEO and digital, who control owned pages and structured data.
- Legal and compliance, who sign off on corrections and disclosures.
- Investor relations, where an inaccurate AI summary can reach the market.
- The subject matter leads whose expertise the earned programme is built on.
Give the owner an executive sponsor and an agreed route for acting on quarterly reports.
What four metrics does this role own?
Four metrics are key to tracking AI visibility, and each is reported per model. Models can disagree, so an average across them can hide the information you need.
- Presence rate: how often the firm is named in answers to the prompts that matter to you.
- Share of voice across models: your presence against a named competitor set, measured per model.
- Accuracy and sentiment: whether what the model says is correct, and how it frames you.
- Source attribution: which third-party pages the model cites when it describes you.
Measure each one over time. A single answer can identify an issue, but repeated sampling is needed to assess how often it occurs.
What tooling is actually needed, and what is optional?
Start with a documented testing method, then add tooling where the scale or reporting requirements justify it. Several vendors track AI mentions. The comparison below focuses on how clearly they explain their measurement methods and limitations. Public disclosure helps you assess a tool, but does not establish its accuracy.
Our review covered 16 vendors, using information published on their own websites and captured on 7 September 2026. We recorded available pricing, platform coverage, explanations of methodology and stated limitations.
The following examples illustrate what vendors disclose and the questions to ask when assessing their tools:
- Ahrefs cautions that the same query can produce different AI answers throughout the day, and recommends tracking visibility over time.
- Evertune illustrates how repeated sampling can narrow the margin of error. Its published example reports roughly ±44 percentage points for a single response and ±10 for 100 repetitions of the same prompt. The ±1-point figure applies to an overall result based on 10,000 responses across a set of prompts, rather than to an individual prompt.
- Meltwater says its platform focuses on directional accuracy and consistent trends rather than individual responses.
- Scrunch AI emphasises that measurement is a representation of reality, rather than a complete account of it.
These disclosures help explain how to interpret the results, but they do not independently establish a tool’s accuracy.
Coverage also matters, particularly for smaller firms. Ahrefs says its pre-collected AI Visibility Index works best for established brands or topics with meaningful search demand, and may offer limited coverage for brands with little or no search volume. That limitation applies specifically to the index, so smaller B2B firms should check whether it covers their brand and relevant topics.
Cision’s AI Visibility Dashboard is powered by Trajaan, which Cision acquired in December 2025. The dashboard launched in July 2026.
| Vendor | Platform coverage | Disclosed methodology | Admitted limitation | Candour |
|---|---|---|---|---|
| Evertune | Major consumer models | Published margin-of-error figures | Confidence depends on repetition volume | Highest |
| Ahrefs | AI Overviews and chat models | Admits non-determinism in public | Thin coverage for low-search-volume brands | High |
| Meltwater | Multiple models | States directional-accuracy focus | Avoids single-response claims | Medium |
| Scrunch AI | Multiple models | Frames measurement as a model | Makes no ground-truth claim | Medium |
| Cision | Dashboard via Trajaan | Limited public methodology | Method undisclosed | Low |
Table captured 7 September 2026. Scheduled for quarterly review. We use these tools on occasion for clients but we have no partnership relationship with any vendor listed, and we resell none of them.
What you actually need is one tool that reports per model and over time, plus a documented method you can defend to a sceptical board. Wall-to-wall platform coverage and real-time dashboards are optional.
What should the first 90 days cover?
The first 90 days build a baseline and a reporting habit. Resist the urge to buy tooling in week one.
- Weeks 1 to 2: agree on the prompt set that matters to your buyers and your board, and name the competitor set.
- Weeks 3 to 4: capture a baseline across the main models, and record which sources they cite.
- Weeks 5 to 8: fix the clearest factual errors at source, and brief the earned programme towards the pages models trust.
- Weeks 9 to 12: publish the first quarterly report on the four metrics, and set the review cadence.
By the end of the first quarter, the owner should have a documented baseline, an issue log and an agreed reporting cycle. That is the point at which the role stops being a project and becomes a function.
How do you make the case to your executive team?
Lead with risk and reach, and keep it to one page. The argument is that AI systems now sit between your firm and the people forming an opinion of it, and that someone needs to be responsible for monitoring the firm’s representation and coordinating a response to significant errors.
Show the executive team examples of how AI systems describe your firm, highlighting material inaccuracies or gaps in relevant answers. Explain why those findings matter to clients, prospective clients or other stakeholders, while being clear about the limits of the testing.
Set out who would coordinate the work, which teams would contribute and what resources are needed. Propose an initial 90-day remit covering a baseline assessment, a process for addressing significant errors and a regular reporting cycle.
Frequently asked questions
Who should own AI search visibility in a company?
Communications should own it, reporting to the head of communications or to the CMO where communications sits inside marketing. A lot of the sources AI models cite are earned, so the owner is the team that already manages reputation and media. Give the role a named executive sponsor.
Is GEO a marketing or an SEO job?
It is a communications job that works closely with SEO. SEO controls your owned pages and technical health, which feed the model. Third-party coverage are also used by AI models, further pushing the role into communications’ remit.
What metrics measure AI visibility?
Four metrics measure it: presence rate, share of voice across models, accuracy and sentiment, and source attribution. Report each one per model on a regular cadence.
What tools track brand mentions in AI?
Vendors in this category include Evertune, Ahrefs, Meltwater, Scrunch AI and Cision, among the 16 we verified at source on 7 September 2026. Evertune and Ahrefs publish the most about their method, including confidence figures and an admission of non-determinism. Pick one that reports per model and over time, and keep the comparison dated.
