The GEO Directory · 19 November 2026 · Published
GTM engineer, marketing engineer, or content engineer? Start with the broken handoff.
A founder who needs AI-search visibility work done will find three job titles that claim it. The postings disagree about what the role is. The backlog does not care what you call the owner.
THREE POSTINGS / ONE PROBLEM
The same work,
three business cards.
A founder reads three job ads and concludes the market cannot decide who owns AI-search visibility. That reading is correct, and it is also useless. The titles are not converging because the work is not one job. It is a set of tasks with different evidence requirements, and they fail at the handoffs between people, not inside any one person's job description.
This guide takes a different starting point. Read what three real employers actually assigned, borrow the vocabulary the discipline already built, then find where your own work stalls. Only then decide who owns what, and write down what "done" means for each task before anyone starts.
01 / Read the postings
Three employers, three different boundaries
Profound's Marketing Engineer builds the engineering layer beneath the marketing team: agents, data pipelines, citation tracking, structured data, publishing integrations. The posting is explicit that this is not a writing job. [1]
Greenly's AI Content Engineer runs the visibility program itself: the GEO audit, the citation-share baseline, AI-assisted article production with quality guardrails, and a video pipeline. When the audit finds a technical blocker, a separate web developer implements it. The title says content; the boundary says strategy and process, with engineering support on request. [2]
August's GTM Engineer is a software engineer whose subject is acquisition. The search-relevant slice covers technical SEO automation, sitemap and schema generation, Search Console API integration and experiment frameworks. [3]
Profound · New York, on-site
Marketing Engineer
$135,000–$200,000 base, as published
Builds the agents, data infrastructure and publishing integrations beneath the marketing team: citation tracking, structured data at scale, CMS and distribution plumbing, experiments. The posting says this person is not writing blog posts. [1]
Greenly · Paris, hybrid
AI Content Engineer
€50,000–€60,000, as published
Owns the GEO strategy and audits, tracks citation share against competitors, scales article production with AI workflows under quality guardrails, and turns articles into video. Technical blockers go to a separate web developer who implements them. [2]
August · New York, on-site
GTM Engineer
$140,000–$220,000, as published
A software engineer for acquisition infrastructure. The search-relevant part: technical SEO automation, sitemap and schema generation, Search Console API integration and experiment frameworks. Its broader outbound systems are outside this article’s scope. [3]
Three postings read on 18 November 2026, quoted as snapshots of what these three companies asked for. Salary ranges are as published and differ by market and company; they say nothing about what any other employer pays. Job postings change or close; the scopes, not the vacancies, are the evidence here.
02 / Name the work
Four practices, whatever the title says
The practitioner vocabulary already exists. Ahrefs describes content engineering as four overlapping practices: pipeline design, skills and prompts, knowledge and source of truth, and orchestration with governance. Its analysis of twenty US content-engineering postings found pipeline building more common than writing itself. Treat that as one vendor's small sample, not a census, but the four practices are a usable checklist. [4]
AirOps proposes a team structure with the same skeleton at org scale: someone maintains the shared context everyone trusts, someone builds the workflows, someone turns performance signals into the next bets, and an executive clears the path. At a startup, those four jobs can be one person on different days of the week. The layers still exist whether or not they have owners. [5]
Pipeline design
Break research, drafting, optimization, publishing and measurement into discrete steps that can run as a system instead of starting from a blank page each time.
Skills and prompts
Turn one senior editor’s decisions into reusable instructions the whole team’s pipeline applies every time it runs.
Knowledge and source of truth
Keep product facts, positioning and proprietary research structured and connected, so generated output draws on company knowledge instead of generic filler.
Orchestration and governance
Schedule and trigger the work, then hold it to fact-checking, citation verification, brand-voice rules and human review checkpoints.
Practice names as published by Ahrefs. Both Ahrefs and AirOps sell tooling for this work; their frameworks are practitioner viewpoints to compare against your backlog, not neutral research.
03 / Find the break
Work stalls at the handoff, not the hire
An engineering-operations framework published by Norveliq makes the useful point: content operations fail like production systems fail. Late approvals, unclear ownership, inconsistent briefs, pages that ship without checks, and a backlog that grows faster than the team repairs it. The remedy is borrowed from engineering: a state model instead of a calendar, structured briefs as interfaces between people, validation gates before publication, throughput and rework measured separately, and a repair path for everything already live. [6]
That framework is prior work, not something this article invented. The contribution here is narrower: apply it to the role question. When a founder asks whether to hire a GTM engineer, the diagnostic question is which state in the workflow has no owner. If approved topics pile up waiting for someone who can change the template, the gap is implementation. If pages ship with claims nobody verified, the gap is evidence. If reports disagree and nobody can say why, the gap is measurement. Three different gaps, three different hires, one misleading title debate.
04 / Run the exercise
Ten tickets, each with one accountable owner
Here is a hypothetical backlog for a startup whose buyers find it through search and AI answers. It is invented for teaching: not customer data, not a benchmark, not a description of any real company. The discipline is the columns. Every ticket names one accountable owner by capability rather than title, the evidence the work requires, and the condition that closes it. Titles can attach to the owners afterward.
| Ticket | Accountable owner | Required evidence | Acceptance condition |
|---|---|---|---|
| 01A stale product claim circulates in a reusable comparison summarySupport: Product marketing | Content owner | Current claim text, the correct claim, and the dated source that settles it | The summary matches the current product page and its refresh date is recorded |
| 02A comparison page makes claims without sourcesSupport: Analyst or researcher | Content owner | A primary source URL for every numeric or factual claim | Every claim carries a dated primary source, or the claim is removed |
| 03A crawler appears unable to reach a public pageSupport: Search/SEO owner | Web developer or technical owner | A verified request where possible, or a synthetic test explicitly labeled synthetic, with the response kept | The delivered body contains the product facts for the intended crawler, or the exact blocking layer is named |
| 04A page family ships with the wrong canonical or templateSupport: Search/SEO owner | Web developer | Rendered HTML of the canonical tag across the family | Every page in the family points its canonical at the intended URL |
| 05A merged change never reaches the live siteSupport: Content owner | Engineering | Deployment identifier plus the final production URL | The final URL returns the new content on production, checked on the live host |
| 06A promotional price spreads into reusable summaries uncontrolledSupport: The person who approved the price | Content owner | The approved price source and its approval date | Every reusable summary shows the approved price with its review date, or marks it unknown |
| 07Analytics include the team’s own testing trafficSupport: Engineering | Analytics owner | A production-host report with the explicit test exclusions applied | A recent sample shows QA traffic excluded by rule, not by assumption |
| 08Nobody can count sign-ups that came from searchSupport: Engineering | Analytics owner | The conversion event definition and a production record of it firing | A real path from search visit to sign-up records the event end to end |
| 09Two AI visibility reports use different question setsSupport: Analyst | Whoever runs visibility tracking | The question panel definition and the date it changed | Comparisons use the same questions both times, and any panel change is recorded with its date |
| 10Traffic grows but buyers never ask the questions the pages answerSupport: Content owner | Growth lead or founder | A list of actual buyer questions mapped to pages | Every published page maps to a named buyer question; unmapped work is deprioritized |
- 01
A stale product claim circulates in a reusable comparison summary
- Accountable owner
- Content owner
- Support
- Product marketing
- Required evidence
- Current claim text, the correct claim, and the dated source that settles it
- Acceptance condition
- The summary matches the current product page and its refresh date is recorded
- 02
A comparison page makes claims without sources
- Accountable owner
- Content owner
- Support
- Analyst or researcher
- Required evidence
- A primary source URL for every numeric or factual claim
- Acceptance condition
- Every claim carries a dated primary source, or the claim is removed
- 03
A crawler appears unable to reach a public page
- Accountable owner
- Web developer or technical owner
- Support
- Search/SEO owner
- Required evidence
- A verified request where possible, or a synthetic test explicitly labeled synthetic, with the response kept
- Acceptance condition
- The delivered body contains the product facts for the intended crawler, or the exact blocking layer is named
- 04
A page family ships with the wrong canonical or template
- Accountable owner
- Web developer
- Support
- Search/SEO owner
- Required evidence
- Rendered HTML of the canonical tag across the family
- Acceptance condition
- Every page in the family points its canonical at the intended URL
- 05
A merged change never reaches the live site
- Accountable owner
- Engineering
- Support
- Content owner
- Required evidence
- Deployment identifier plus the final production URL
- Acceptance condition
- The final URL returns the new content on production, checked on the live host
- 06
A promotional price spreads into reusable summaries uncontrolled
- Accountable owner
- Content owner
- Support
- The person who approved the price
- Required evidence
- The approved price source and its approval date
- Acceptance condition
- Every reusable summary shows the approved price with its review date, or marks it unknown
- 07
Analytics include the team’s own testing traffic
- Accountable owner
- Analytics owner
- Support
- Engineering
- Required evidence
- A production-host report with the explicit test exclusions applied
- Acceptance condition
- A recent sample shows QA traffic excluded by rule, not by assumption
- 08
Nobody can count sign-ups that came from search
- Accountable owner
- Analytics owner
- Support
- Engineering
- Required evidence
- The conversion event definition and a production record of it firing
- Acceptance condition
- A real path from search visit to sign-up records the event end to end
- 09
Two AI visibility reports use different question sets
- Accountable owner
- Whoever runs visibility tracking
- Support
- Analyst
- Required evidence
- The question panel definition and the date it changed
- Acceptance condition
- Comparisons use the same questions both times, and any panel change is recorded with its date
- 10
Traffic grows but buyers never ask the questions the pages answer
- Accountable owner
- Growth lead or founder
- Support
- Content owner
- Required evidence
- A list of actual buyer questions mapped to pages
- Acceptance condition
- Every published page maps to a named buyer question; unmapped work is deprioritized
Hypothetical teaching backlog. Some tickets are engineering work, some are editorial, some are analytical. Notice that no single one of the three job postings from section 01 would obviously own all ten.
05 / Measure honestly
Know what the report can and cannot tell an owner
Every owner in the exercise needs measurement they can defend. Google Search Console now has a dedicated generative AI performance report. The current documentation says it rolled out to all websites worldwide on 31 August 2026, following the June 2026 announcement. [7] [8]
Read its limits before assigning anyone a target. The report covers impressions of links to your site inside AI Overviews and AI Mode. You can group by page, country, date and device, and filter text or multimodal searches. It does not show the query or prompt text. It reports impressions, not clicks or conversions, and its table shares the usual Search Console row limits. A report that never appears can mean too few impressions, or that the site is excluded from Search generative AI features, which is a separate control. [7]
Eligibility is the other half. Google's AI features documentation ties inclusion to foundational Search eligibility and visible content, and guarantees nothing: a page can be eligible, indexed and still never surface in an AI feature. Assign owners to evidence and acceptance conditions they control, not to outcomes no control panel promises. [9]
What the owner can be held to
Evidence and acceptanceDated sources, verified delivery, same-question comparisons, recorded panel changesWhat no owner can promise
Citations and answersImpressions where Google reports them, nothing where it does not, no guaranteed inclusion anywhere06 / Decide
One person can hold all three scopes, until a handoff fails
Early on, one generalist can legitimately cover the strategy, the pipeline and the measurement. The postings show companies splitting the same work three ways; nothing says a five-person startup should. The split becomes necessary when a specific handoff fails repeatedly: facts nobody maintains, work stuck between brief and publish, numbers nobody trusts. Each failure maps to one of the four practices in section 02, and each maps to a capability you can hire, contract or deliberately keep in-house.
Whatever the structure, some tasks stay human regardless of automation: verifying sources, choosing which buyer questions matter, and signing off the acceptance condition. The employers in section 01 agree on that much. Greenly puts expert review on regulated topics inside its guardrails, and Profound asks for people who challenge assumptions with evidence. Automation carries the load; accountability stays named, dated and auditable. [1] [2]
Pick one stuck task from your own backlog. Write the owner, the evidence it needs and the condition that closes it. If you cannot, the problem was never the missing title. It was the missing sentence.
Method and sources
What is established, and what is not
This is a documentation-led synthesis, not a labor-market study, a hiring recommendation or a product test. Three employer postings establish what three companies asked for on the day they were read. They do not establish a standard role taxonomy, typical pay or that any opening is still available. The Ahrefs and AirOps frameworks are vendor viewpoints from companies that sell into this workflow; the operations framework is prior independent practitioner work, credited rather than reinvented. The ten-ticket backlog is hypothetical and labeled as such.
No search-volume claims are made for these role queries. Suggestion checks from one English-language client on 18 November 2026 showed job and salary intent dominating role queries, and no suggestions for one role-plus-SEO phrase. Suggestions are not volumes, and one client is not the market.
Sources were read on 18 and 19 November 2026. Postings, documentation and reporting features change; check the current pages before repeating any claim here.
- Profound: Marketing Engineer job posting
Employer posting read on 18 November 2026. One company’s scope, not a standard definition.
- Greenly: AI Content Engineer job posting
Employer posting read on 18 November 2026. Targets stated there are aspirations, not results.
- August: GTM Engineer job posting
Employer posting read on 18 November 2026. Only its search and measurement scope is discussed here.
- Ahrefs: What Is Content Engineering, and How Do You Do It?
Vendor practitioner editorial, 19 May 2026. Its 20-posting sample is a vendor analysis, not representative labor-market data.
- AirOps: It’s Time for the Modern Content Engineering Org
Vendor organizational proposal, 2 October 2025. A framework to compare, not evidence of citation gains.
- Norveliq: How to Run SEO Content Operations Like an Engineering System
Independent practitioner framework: states, interfaces, validation gates and repair paths. Publication date not established.
- Google Search Console Help: Generative AI performance report
Current reporting documentation. Rolled out to all websites worldwide as of 31 August 2026.
- Google Search Central: Introducing Search generative AI performance reports
3 June 2026 announcement with a 31 August 2026 availability update.
- Google Search Central: AI features and your website
Technical eligibility and controls, last updated 10 December 2025. Use alongside the newer reporting documentation above.
The GEO Directory. Ownership is a sentence you can audit: a name, the evidence, and the condition that closes the work.