Short answer, as of August 2026: put one named owner on the outcome, borrow the hands. The work spans content, engineering and local-data operations, and no single existing role covers it — so the failure mode is not incompetence, it is diffusion. One accountable person with a standing claim on a few engineering hours a month beats a committee every time.
Most of the programmes that stall do so at a handoff, not at a task. The tasks are well understood by now; who does them, in what order, and who notices when they stop, is not.
Why this falls between the chairs
Look at what a working programme actually requires and the problem is obvious.
Measurement and prompt selection are marketing work. Crawler access, rendering and structured data are engineering work. Business profile, directories and review responses are local-operations work, often sitting with whoever manages the storefront or the front desk. Content upgrades belong to whoever writes.
Each of those groups has a backlog, a manager and a set of metrics — and none of those metrics is "are we named when a buyer asks an assistant about our category". So the work is genuinely nobody's, and the parts that require crossing a boundary are the parts that never get done. That is why crawler-access failures survive for months: the marketer cannot fix a WAF rule and the engineer was never told it mattered.
The three roles, and what each actually contributes
| Role | Owns | Time commitment | Fails when |
|---|---|---|---|
| Visibility owner (marketing) | The outcome, the prompt set, the report, prioritisation | 4–8 hrs/week | Given the metric but no engineering access |
| Technical partner (engineering) | Crawler access, rendering, structured data, migrations | 2–4 hrs/month | Pulled in only after something breaks |
| Content hand (writer or agency) | Page upgrades, new answers, fact maintenance | Variable by phase | Briefed on keywords rather than questions |
| Local-data hand (ops) | Business profile, directories, reviews, NAP | 1–2 hrs/month | Treated as a one-off setup task |
The visibility owner is the non-negotiable one. Everything else can be borrowed, contracted or automated; accountability cannot. If you take one thing from this piece, it is that naming this person is the intervention with the highest return, and it costs nothing.
The second most valuable move is smaller than it looks: a standing, pre-agreed claim on a couple of engineering hours a month. Not a project, not a ticket queue — a recurring slot. Almost every access and rendering failure is a thirty-minute fix that waits eleven weeks for prioritisation.
The four handoffs that break
Marketing to engineering, on access. The visibility owner discovers assistant crawlers are getting a 403 and files a ticket that competes with feature work. It is small, unglamorous and urgent-sounding in a way that engineering has learned to discount. Fix: classify access regressions as incidents rather than requests, with the same standing as a broken form. The diagnostic itself is in the AI crawler access audit.
Engineering to marketing, on releases. A redesign ships and the passage that was earning citations quietly disappears. Nobody did anything wrong; the information simply never travelled. Fix: a protected-content list, and a diff of cited pages as a release checklist item — the discipline described in protecting AI visibility through a site migration.
Marketing to content, on briefs. The writer receives a keyword and a word count and produces something that ranks and never gets cited, because nothing in the brief asked for the specific, verifiable claims that earn selection. Fix: brief the question and the required specifics — the number, the range, the named process. Scoring a draft before you publish it is the check that catches this before it ships.
Everyone to nobody, on maintenance. Citations decay silently and no one's role covers noticing. Fix: put cited-page checks on a calendar and give them to the visibility owner, per citation decay.
A useful test of whether your handoffs work: pick the last access or rendering defect you found and count the days between it starting and someone acting on it. If the answer is measured in months, the problem is the routing, not the people. Teams that shorten that interval do it with two unglamorous mechanisms — a named owner who looks at one chart every week, and a rule that says access regressions skip the backlog.
What one person can realistically hold
For a single-site business, this is a part-time responsibility, not a role — roughly half a day a week once the setup phase is done, plus the borrowed engineering time.
A workable weekly rhythm for that person: review the citation chart for movement, read the answers where a competitor is named instead of you, queue one page upgrade, and check nothing in last week's release touched a cited page. Monthly: the maintenance pass on top cited pages. Quarterly: revisit the prompt set and re-baseline.
What does not fit into half a day a week is the initial setup — building the prompt set, establishing the baseline, running the access audit and fixing whatever it finds. Budget that as a project of two to four weeks, then hand the steady state to the part-time owner. The phased version is in the first 30 days playbook, which reads equally well as an in-house plan.
Multi-location and multi-brand operations are where a part-time owner stops being sufficient, mainly because the local-data surface multiplies faster than anything else — see the multi-location SEO playbook. Ten locations do not mean ten times the content work, but they do mean ten profiles, ten sets of directory records and ten opportunities for the entity data to drift out of agreement, and that portion of the job scales close to linearly.
One more scoping note worth setting expectations on internally: the setup project produces a finding, not a result. Its output is a baseline and a list of blockers, and the natural reaction from anyone expecting a campaign is that nothing visible happened. Saying this in advance — that month one buys knowledge and eligibility, and the visible movement arrives around week twelve — prevents the programme being judged at the point it is least able to show anything.
In-house, agency, or both
The split that works most often is: agency for setup and technique, in-house for the standing ownership. Agencies bring calibration — they have seen what a normal citation rate looks like across categories, which is exactly the judgement a first-timer lacks. What they cannot supply is the internal standing to get a firewall rule changed on Tuesday.
The reverse arrangement, where the agency holds accountability and the client provides no internal owner, is the one that reliably disappoints. Every technical fix routes through someone with no stake in the outcome, and the programme moves at the speed of the least interested party.
If you are choosing a supplier for the setup half, how to choose a GEO specialist covers the questions that separate the calibrated from the confident, and DigiRank vs BrightLocal vs an agency frames the three-way build-or-buy decision.
Give the owner something to own
Accountability without instrumentation is just blame. The visibility owner needs a number that moves, visible to the people who allocate the engineering hours, or the standing claim on those hours will quietly lapse.
DigiRank's AI Visibility Tracker provides that number — citation rate and share of voice per engine, on a schedule, alongside Search Console and GA4 so the marketing report and the engineering justification are the same chart. Access and index monitoring runs continuously in the site audit module, which turns "the crawler has been blocked since March" from an archaeology finding into an alert. Tracking starts on the $99/mo Starter plan for a single brand; multi-tenant, white-label reporting and the site audit and auto-fix pipeline arrive with Agency at $249/mo on a 14-day trial, and in-house teams running several brands usually land on Pro at $499/mo. What to put in the monthly report is covered in white-label SEO reporting for agencies — the format works internally too.
Frequently asked questions
Who should own AI search visibility in a company? One named person in marketing, accountable for the outcome, with a standing claim on a few engineering hours a month and access to whoever manages the business profile and directories. The specific title matters far less than the fact that a single person is accountable rather than a committee.
How much time does this take in-house? After a two-to-four-week setup project, roughly half a day a week for a single-site business, plus two to four engineering hours a month and an hour or two of local-data work. Multi-location operations need more, mostly because the directory and profile surface grows faster than anything else.
Why do AI visibility programmes stall? Almost always at a handoff rather than a task. Access fixes wait in an engineering backlog, releases silently remove cited content, briefs ask for keywords instead of specific answers, and nobody's job includes noticing that a citation disappeared.
Do I need a dedicated hire? Not usually for a single site. This is a part-time responsibility layered onto an existing marketing role, provided that role is given the engineering access and the mandate. A dedicated hire starts to make sense across multiple brands or many locations.
Should engineering own this instead? No. Engineering owns a critical subset — access, rendering, structured data, migration safety — but the prioritisation and measurement are marketing judgements. Handing the outcome to engineering tends to produce a technically clean site that still answers nobody's question.
What is the single highest-return organisational change? Naming the owner, followed closely by giving them a recurring claim on a small amount of engineering time. Most access and rendering failures are thirty-minute fixes that spend months waiting for prioritisation.
How do I get engineering to take crawler access seriously? Classify an access regression as an incident rather than a request, and show the citation chart falling. A metric that moves visibly, in front of the people who allocate the time, does more than any amount of explanation.
Is it better to run this in-house or hire an agency? Most often both: an agency for setup and calibration, an in-house owner for standing accountability. The arrangement that reliably disappoints is an agency holding accountability with no internal owner, because every fix then routes through someone with no stake in the result.