The Job Req You Can't Write
An Executive Sends You an AI Answer, and Someone Scopes a New Role
If you run marketing, you have already received this screenshot or you will. An executive types your category into an AI assistant, reads an answer naming three companies that are not yours, and forwards the image with one line: we are not appearing in LLMs. An LLM is a large language model — the system behind ChatGPT, Gemini, Claude and Perplexity (and, increasingly, the thing your buyer asks before they ever reach your site). What happens next is unusually consistent. Someone scopes a new role.
Before scoping it, two numbers.
On the same 106-Prompt panel, pulled the same day, our brand held 43.1% share of voice on Perplexity and 6.7% on Gemini (Indexable Brand Radar, 2026). Same brand, same questions, same day — a 36-point spread with nothing changed but the model answering. And across 29 public AI-visibility job postings, no two shared a job title, while the median posting asked one hire to cover eight distinct specialisms (Indexable, 2026).
The screenshot was right that something is wrong. It could not say how big, in either direction. And the fix it triggered asks one person to be eight people.
Why Does One Screenshot Turn Into a Headcount Req?
Because the req is the only artifact that is presentable by Friday.
An executive has handed you an unbounded threat with no denominator. You cannot answer it with a measurement plan, because a measurement plan is a promise to have an answer later, and later is not what was asked for. A headcount req converts that unbounded threat into something the board already knows how to read: a named owner, a start date, a budget line, a function that exists.
I want to be clear that this is rational. If I were sitting in that seat with that screenshot in my inbox, I would want the req too. It is the fastest available way to demonstrate that the problem has been received and given to someone. The instinct is sound. What follows is a question about the shape of the thing the instinct produces, not about the instinct.
Because the req has to describe a job. And that is where it comes apart.
What Does One Engine Actually Tell You?
The screenshot measured which engine the executive happened to open.
"Invisible in AI" reads like a state of the world. It is not. It is a per-engine coordinate. Here is our own brand, on our own tracked panel, every figure pulled the same day:
| Engine | Us | The rival we track |
|---|---|---|
| Perplexity | 43.1% | 38.7% |
| Google AI Overviews | 25.5% | 40.7% |
| ChatGPT | 9.6% | 83.8% |
| Gemini | 6.7% | 43.5% |
Indexable Brand Radar, n = 106 Prompts, US, 31 Aug 2026. Per-brand share of voice on our own panel — they do not sum to 100%. Rival anonymised.
Yes, we are the 6.7%. That row is why this section exists rather than a tidier one. In July we published ourselves as the leader on Google AI Overviews. We re-pulled before posting the follow-up and found we no longer were — and when we re-pulled again on the morning this piece went out, all four of these numbers had moved, ours and theirs, in both directions. The rival's Gemini share had nearly halved in a fortnight. We publish the range and the method for that reason: Citation Volatility covers how to measure the spread, and this piece does not re-teach it. The prior episode, Nine Page-One Rankings, Zero Clicks, covers why the traffic report will not flag any of this.
The relevant point for the req is narrow. Had that executive opened a different assistant, the screenshot would have said something else, and the role scoped in response would have had a different shape. One engine cannot tell you how big the problem is. It can only tell you that someone looked.
What Are These Reqs Actually Asking One Person to Do?
This is the part the req cannot survive.
I collected 29 public job postings in the AI-visibility, GEO and AEO space from public job boards between June and August 2026, and counted the distinct skill domains each posting demands. GEO and AEO are generative and answer engine optimisation (the work of getting cited inside an AI answer, as distinct from classic SEO, which is a discipline aimed at ranking a link on a results page). You can read the difference as: one earns the citation, the other earns the click.
Not the buzzwords — the domains: technical SEO, structured data, content strategy, content production, analytics and measurement, engineering and automation, link acquisition, paid, CRO, localisation, people leadership, cross-functional influence, product and roadmap work.
Across 29 postings there were 29 distinct job titles. Not one repeated. Seniority was scattered across the same scope of work — eleven scoped as Head or Lead, eight as Manager, seven at Director level or above, three as individual contributors (Indexable, 2026). The same job is being posted simultaneously as an IC role and as a Director role. The market has not agreed what this is called or what level it sits at.
The median posting asked for eight distinct specialisms. The range ran from four to twelve. Seventeen of the 29 asked for eight or more. Of those seventeen, ten asked for no people leadership at all (Indexable, 2026). No reports. No team. No delegation. One person, personally, across eight domains.
Which Specialisms Are Actually Being Demanded?
Not the ones you would expect.
Analytics and measurement appeared in 86% of postings. Engineering or automation appeared in 86%. Both outrank content strategy and content production, which appeared in 59% each (Indexable, 2026).
The role being scoped is not a content role with AI bolted onto it. It is a measurement-and-systems role that also has to write well. That combination has always been rare, and it is not about to become less rare on a hiring timeline.
Thirty-eight percent wanted technical SEO and content production and analytics and engineering automation from the same hire. In any other marketing org, those are four people.
Sixteen of the 29 disclosed a salary band. The median ran $175,600 to $221,175 in base pay alone (Indexable, 2026). That is the going rate for one person to be eight people.
One employer in the corpus did something different. Faced with the same function, they opened four simultaneous requisitions instead of one — splitting growth, performance, partnerships and product marketing rather than compressing them. Their remaining organic role still asked for eight specialisms. Eight was what was left after three other jobs had been carved out of it.
What Can a Fleet Genuinely Not Do?
Three things, and they are worth naming plainly.
When nobody owns organic demand at all, the first hire is org design, not execution. If there is no one in the building whose job it is to care about this surface, buying capability adds output to an organisation with nowhere to put it. Start by hiring the person. Buy nothing until they exist.
When the bottleneck is permission rather than production, more production does not help. If the real constraint is that nobody can get a sprint slot from engineering or a decision out of legal, the answer is a human with standing in those rooms. No agent negotiates a sprint slot. You can test which constraint you have in an afternoon: ask what shipped last quarter, then ask what was written and never shipped. If the second list is longer, you have a permission problem.
In regulated and claims-heavy categories, every published sentence needs a named human reviewer. Finance, health, legal, insurance. That review is not a bottleneck to be automated away; it is the control that keeps the category legal.
If your situation is one of those three, hire, and do not let anyone — including me — talk you out of it.
So What Should the Req Say Instead?
Keep the Friday artifact. Change its shape.
You still need the named owner, the start date and the budget line. Those are what make the problem legible to a board. What changes is what the owner is asked to personally do.
Start by cutting the specialism list out of the responsibilities section entirely. Then, scope the role like this:
- Owns the surface. Accountable for how the brand appears across search and AI answers — the outcome, not the task list.
- Sets direction. Decides which questions matter, which engines matter, and in what order.
- Holds the room. Carries standing with engineering, legal and product. This is the part no system does for you.
- Directs execution across the specialisms rather than personally performing eight of them.
- Is measured on coverage and outcome, not on how many of the domains they can personally operate.
The version that cannot be written is the one that answers "all of them." That is the req you have been handed, and it is the reason the search keeps failing.
This one is writable. It describes a job a person can hold, and it survives the question a good candidate asks in the first interview: which of these eight am I accountable for personally?
Hire the strategist. Give them the fleet.
That is what Indexable is built as — Enterprise AI SEO Agents run as one coordinated team, with a forward-deployed strategist alongside them.
Frequently Asked Questions
Is this an argument against hiring?
No. It is an argument against a specific req shape — one person asked to personally perform eight specialisms with no team. The strategist hire is the right call; the corpus suggests the scope written underneath it usually is not.
Why does the number of specialisms matter more than the number of hires?
Because a req describes what one person will personally do. When the median posting names eight distinct skill domains and most name no reports, the org has decided the work is one job before checking whether anyone can hold it. The count is how you check.
How was the corpus collected?
29 public postings in the AI-visibility, GEO and AEO space, captured from public job boards between June and August 2026 and coded for the distinct skill domains each demands. It is a convenience sample, not a random one — it skews senior, US and B2B software, so treat the counts as indicative of that segment rather than of the whole market. Employers are aggregated and not named.
Our AI visibility looks fine on the engine we checked. Are we done?
That is the finding this piece is least comfortable with. On our own panel our share ranged from 43.1% to 6.7% on the same day depending only on which model answered (Indexable Brand Radar, 2026). Checking one engine tells you about that engine. You should check the spread before you scope anything against it.
Does this apply outside B2B software?
The corpus is B2B-weighted (business-to-business software companies hiring for their own marketing teams), so treat the specific counts as indicative there. The structural point — that one screenshot produces one req, and the req names more domains than one person holds — is not category-specific.
Key takeaway: the screenshot is a real signal and a bad measurement. Across 29 public postings there were 29 different titles and a median of eight specialisms per role, most with nobody to delegate to (Indexable, 2026). The next step is not a longer responsibilities list. It is a shorter one, with coverage underneath it.
Written by Vijay Vasu, founder of Indexable AI, who has led organic growth at Zendesk, Uber Eats and Williams-Sonoma. Figures are first-party unless otherwise sourced. Data: Indexable Brand Radar, n = 106 Prompts, US, pulled 31 August 2026; job-req corpus, n = 29 public postings collected June–August 2026.