TL;DR: Most teams buy an LLM SEO tool, get a dashboard, and publish nothing differently six months later. The tool is rarely the problem. Measurement costs a subscription and acting on it costs 25 to 70 hours a month, and almost nobody budgets for the second number. Decide who will do the work before you decide which tool to buy, because the tool is the cheap part.
Key Takeaways
Tools diagnose; they do not ship. Every finding becomes a task for a person, and the task list is longer than the dashboard suggests.
The hours are the real cost. Subscriptions run roughly 3% to 15% of what a programme actually costs once execution labour is counted.
Findings route to four different owners. Content, engineering, brand and whoever manages third-party profiles. Only one of those usually reports to the person who bought the tool.
A dashboard nobody acts on is worse than nothing. It creates the impression of a programme and consumes budget that could have funded the work.
Three routes close the gap. In-house capacity, an agency, or a managed service. Each has a shape that suits it, and doing nothing is the common fourth.
Decide the route before the tool. If you cannot name who ships the fixes, buying measurement is premature.
Anchor targets assume your CMS generates slug IDs from heading text. Confirm on staging and adjust the hrefs if your platform slugifies differently.
What these tools do
An LLM SEO tool, also sold as an AI visibility tracker or AI search monitor, runs a list of prompts against AI assistants on a schedule and reports whether your brand was named and whether your URLs were cited.
That is genuinely useful. It is also the entire scope, and being clear about the boundary is what prevents the disappointment that follows most purchases.
What you get
Capability | Value |
|---|---|
Mention and citation rates across your prompt set | A baseline and a trend |
Which competitors appear instead | Your real comparison set |
Which URLs were cited | The most actionable output |
Full answer text, in better tools | How you were described, and why you lost |
Trend over time | Whether the work is paying |
What you do not get
Not included | Who has to do it |
|---|---|
Pages written against the gaps found | Content |
Openers rewritten to be extractable | Content |
Crawler access fixed | Engineering |
Rendering changed so content is readable | Engineering |
Schema added or corrected | Engineering |
Third-party profiles corrected | Whoever owns them, often nobody |
Entity details made consistent across sources | Marketing ops |
Seven kinds of work, none of them performed by the subscription.
A note on what cannot be done at all
Some claims attached to this category describe things no tool or service can do, and they are worth separating from the real gap.
Nothing weaves your brand into training datasets. Training corpora are assembled by model providers on their own schedule and no third party edits them. Nothing measures or tunes your position in a commercial model's vector space, because that is not externally accessible. What you influence is retrieval: whether your pages can be reached, whether they contain quotable passages, and what third-party sources say about you.
That is a smaller claim and it is the true one, and everything in this article concerns closing the gap between knowing those things and changing them.
The execution gap
The gap is the distance between a finding and a shipped change. It is measured in hours and it is far larger than the subscription.
The arithmetic
A tool reports that you are cited on 23% of your prompts. Useful. Now consider what closing even part of that gap requires monthly.
Work | Hours |
|---|---|
Weekly triage of lost prompts | 2 to 4 |
Writing pages against identified gaps | 12 to 40 |
Rewriting openers on existing pages | 4 to 8 |
Technical fixes, specified and shipped | 2 to 10 |
Third-party profile and entity corrections | 2 to 6 |
Reporting and stakeholder management | 2 to 4 |
Total | 24 to 72 |
These are practitioner estimates and they scale with how much you intend to fix. The point is the order of magnitude rather than the precision.
Against the subscription
Line | Monthly |
|---|---|
Tool subscription | $29 to $500 |
Execution at 25 hours, $85 loaded | $2,125 |
Execution at 70 hours, $85 loaded | $5,950 |
The subscription is somewhere between 3% and 15% of the real cost. Teams negotiate hard on the small number and never budget the large one, which is the whole problem in one sentence.
What actually happens instead
The common pattern is recognisable.
Month one: the tool is set up and the baseline is interesting. Month two: a report goes to the marketing lead, who forwards it. Month three: a few openers get rewritten by whoever had time. Month six: the dashboard is open in a tab nobody looks at, the rate has not moved, and the renewal is questioned.
Nothing went wrong technically. The measurement worked. There was simply no capacity attached to it.
The honest test
Before buying, ask: if this tool told me tomorrow that we lose forty prompts because we have no page answering them, who writes those forty pages, and when?
If the answer is vague, the gap is already there and the subscription will not close it. That question is worth more than any feature comparison.
Why a dashboard does not produce work
It is worth understanding why this happens so reliably, because the reasons are structural rather than a failure of discipline.
A score is not a task
A citation rate of 23% tells you where you stand. It does not tell anyone what to do on Tuesday.
Converting a rate into a task requires reading the lost prompts, reading the answers that beat you, deciding whether the cause was absence, format or access, and writing a brief. That translation step is skilled work and no dashboard performs it.
Tools that stop at a score have handed you a number and kept the hard part.
The findings belong to other people
The person who buys the tool is usually in marketing. The fixes land across four functions, and only one of them is theirs.
Asking engineering for a rendering change on the strength of a marketing dashboard is a request without authority behind it. It joins a backlog and it waits.
Nobody is measured on it
The content team has a publishing calendar. The engineering team has a roadmap. Neither has citation rate in their objectives, so work generated by the tool competes with work they are accountable for and loses.
This is the most common reason programmes stall and it has nothing to do with the tool's quality.
The work is unglamorous
Much of what closes the gap is dull: correcting a directory listing, rewriting an opening paragraph, checking server logs, making a phone number consistent in six places.
It is also the highest-return work available, and it is nobody's favourite task, which means it needs to be assigned rather than hoped for.
The fix
Three things, all organisational rather than technical.
Fix | What it means |
|---|---|
A named owner with a route across functions | Someone who can get an engineering ticket prioritised |
Committed capacity, in hours | Not spare time. Booked time, in a plan |
The output expressed as tasks with owners | Not a score. A list of named gaps, each assigned |
If you have those three, almost any competent tool will work for you. Without them, the best tool in the category will produce a chart and nothing else.
Four findings, four owners
Every lost prompt has one of four causes. Each needs a different person, and mapping them is what turns a dashboard into a work queue.
The mapping
What you see | Cause | Owner | Effort |
|---|---|---|---|
Nobody cited, question unanswered on your site | No coverage | Content | Hours per page |
You have a page, a competitor's was quoted | Not extractable | Content | 30 to 60 min per page |
Your page exists and never appears | Not retrieved | Engineering | Varies, often one line |
Named but described wrongly | Bad source data | Marketing ops | Minutes to hours |
Why this mapping matters more than the score
Two of the four are cheap. Rewriting an opener takes under an hour and the page is already indexed. Correcting a directory listing takes minutes.
Two are expensive. Writing new pages is the largest cost in any programme, and engineering fixes depend on a backlog you do not control.
A team that sorts findings this way does the cheap work first and gets movement while the expensive work is still being scheduled. A team looking only at a rate has no idea which it is facing.
Check this one first
Not retrieved. If your pages exist and never surface, no amount of writing helps, and the cause is usually a crawler block or client-side rendering rather than anything about your content.
Confirm it in server logs by user agent and status code. That single check has rescued more programmes than any content strategy, it takes about an hour, and it is frequently a one-line configuration fix once found.
Turning the mapping into a routine
Thirty to sixty minutes weekly: pull the prompts you lost, read the answers, sort each into one of the four buckets, and add it to the owner's queue with a link to the evidence.
That routine is the entire difference between a tool that pays for itself and a tool that renews unexamined. It is also the thing most likely to be dropped in a busy month, which is why it needs a named owner rather than good intentions.
Closing the gap in-house
The cheapest route if you have the people, and the one that fails quietly if you do not.
What it requires
Requirement | Detail |
|---|---|
A named owner | One person accountable for the programme, not a committee |
Booked capacity | 25 to 70 hours monthly across functions, in plans rather than spare time |
A route to engineering | Someone who can get a ticket prioritised, or a standing allocation |
Someone who can write | The largest share of the hours |
A weekly routine | The triage, owned and calendared |
The realistic staffing
Volume | What it looks like |
|---|---|
Light, a few pages monthly | A marketer at roughly 25% of their time, plus occasional engineering |
Moderate | A marketer at half time plus a writer, or one dedicated person |
Heavy | A dedicated person plus content capacity, in-house or freelance |
The advantages
Your own people know the product, which matters more here than anywhere else. The specifics that make content citable, which are real constraints, real numbers and genuine opinions, are things only insiders have.
You also keep the capability. Three quarters in, your team understands your category's prompt landscape in a way no external party will.
The failure mode
Capacity that exists on paper and not in practice. Someone is given this alongside a full workload, the triage slips in a busy month, then slips again, and by quarter three the subscription is paying for a dashboard nobody opens.
The tell is simple: if the weekly triage has been missed twice in a row, the capacity is notional and the programme is already stalled.
Make it survivable
Put the triage in a calendar with a named owner. Report shipped changes rather than the score, so the work is visible even in months where the rate does not move. And agree the hours with whoever owns that person's time, in writing, before the tool is bought rather than after.
Closing it with outside help
Two external routes, with different shapes.
An agency
Typically retained for content production, sometimes with technical SEO alongside. You keep the tool and the strategy; they supply capacity.
Works when | Struggles when |
|---|---|
Content volume is your constraint | The constraint is technical and sits with your engineers |
You can brief well | Nobody internally can review for product accuracy |
You want to keep strategy in-house | You wanted someone to own the outcome |
The thing to settle upfront is who does the triage. If you expect the agency to read lost prompts and set priorities, that is a different engagement from one where you hand them briefs.
A managed service
Measurement, diagnosis and execution under one owner, which removes the handoff between knowing and doing. That is the entire argument for the model and it is a reasonable one.
Works when | Struggles when |
|---|---|
You have no capacity and will not hire | You want to build the capability internally |
You want one accountable party | Your constraint is engineering access they do not have either |
Standard requirements | Highly specialised category where product knowledge is everything |
What to check in either case
Question | Why |
|---|---|
Who performs the weekly triage? | The step that converts data into work |
How are technical findings handled? | Nobody external can deploy to your site. They specify; you ship |
What is the committed output volume? | Hours or deliverables, in the contract |
Who reviews for factual accuracy? | Outsiders do not know your product. This stays yours |
What happens to the data if we leave? | Your measurement history is the thing you cannot rebuild |
That second row is the one people miss. No external party can push a rendering change to your site. They produce a specification and your engineering team deploys it, which means the engineering bottleneck persists regardless of who you hire.
The honest limit of outsourcing
Two things never transfer: your product knowledge and your deployment access.
The specifics that make content worth citing come from inside, and the technical fixes ship through your people. Any arrangement that claims otherwise is either overselling or will deliver generic content and unshipped specifications.
Budget internal time even when outsourcing. Less of it, and not none.
The three routes compared
Tool plus in-house | Tool plus agency | Managed service | |
|---|---|---|---|
Monthly cost | Subscription plus internal hours | Subscription plus retainer | One fee |
Internal hours needed | 25 to 70 | 10 to 25 | 5 to 15 |
Who does triage | You | Negotiate it | Them |
Who writes | You | Them, from your briefs | Them |
Who ships technical fixes | You | You | You |
Product knowledge | Yours, the advantage | Needs transferring | Needs transferring |
Capability retained | Yes | Partly | Little |
Accountability | Internal | Split | Single party |
Suits | Teams with capacity | Content-constrained teams with a strategist | Teams with neither |
The engineering row is identical across all three, because nobody external deploys to your site. If technical issues are your main gap, no purchasing decision solves it and the conversation you need is with your engineering lead.
The fourth route
Buy the tool, assign nobody, and do nothing. It is the most common outcome and it is worth naming because it is rarely chosen deliberately. It happens when capacity is assumed rather than booked.
Its cost is the subscription plus the opportunity cost of a quarter, plus the harder-to-measure cost of a team concluding that this channel does not work when it was never actually attempted.
Choosing
If | Then |
|---|---|
You have 25 or more hours a month you can genuinely commit | Tool plus in-house |
You have a strategist but no writers | Tool plus agency |
You have neither, and will not hire | Managed service |
You cannot name who would do any of it | Buy nothing yet |
That last row is a legitimate answer. Spending a quarter building the capacity, or getting the free foundations done, puts you in a better position than a subscription would.
The decision order
Route first, tool second. Almost everyone does this the other way round, which is why the gap is so common.
Once you know who will act on the findings, the tool choice becomes straightforward and comparatively low-stakes: accuracy, full answer text, cost per tracked prompt, and export rights.
What to fix before buying
Some of what a tool would tell you, you can find out for nothing this week. Doing so makes the eventual purchase smaller and better targeted.
The four free checks
1. Run twenty prompts manually. Five runs each, across three assistants, logged in a spreadsheet. Two hours. You get a baseline, and you learn how you are described, which is usually the more interesting half.
2. Check crawler access in your server logs. Filter by AI user agent, look at status codes. One hour. If retrieval crawlers are blocked, that is your whole problem and no subscription would have fixed it faster.
3. Audit your entity consistency. Business name, category, description, address, phone across your site, your profiles and the main directories. A few hours. Inconsistency here is a common reason models hedge about you.
4. Read your top ten pages' opening paragraphs. Does each answer a specific question in its first fifty words, or open with positioning? This one is free and immediate.
What these usually reveal
Finding | Frequency | Fix |
|---|---|---|
Crawler access blocked somewhere | Common | Often one configuration line |
Entity details inconsistent | Very common | Hours of data entry |
Openers are positioning, not answers | Very common | Under an hour per page |
Genuinely no coverage of key questions | Common | Real content work |
Three of four are cheap and none of them needs a tool to find.
Then buy against what is left
If after fixing those your remaining gap is content coverage across many questions, you have a production constraint and now you know its size. If you simply cannot keep checking manually, you have a measurement constraint.
Either way the purchase is now specific, justified and the right size, which is a much better position than buying a dashboard and hoping it tells you something.
The uncomfortable likelihood
For many teams these four checks surface a quarter's worth of work that needs no software at all.
Doing that work first means the tool, when you buy it, measures a site that has already had its obvious problems fixed, rather than spending three months telling you about them.
Six mistakes that create the gap
1. Buying before naming who acts. The single cause of most stalled programmes. If you cannot say who writes the pages a tool will tell you to write, the subscription buys a chart.
2. Budgeting the subscription and not the hours. Execution runs 25 to 70 hours a month, which dwarfs any fee in this category. A plan with one of those numbers in it is not a plan.
3. Treating the score as the deliverable. A rate tells nobody what to do on Tuesday. The deliverable is a list of lost prompts sorted by cause, with owners.
4. Assuming marketing can ship technical fixes. Crawler access and rendering sit with engineering, who have their own roadmap and no citation rate in their objectives. Secure that route before you need it.
5. Expecting an external party to close the whole gap. Nobody outside deploys to your site, and nobody outside knows your product's real constraints. Both of those stay with you regardless of what you buy.
6. Skipping the free checks. Crawler access, entity consistency and page openers can all be checked this week for nothing, and they are where the problem usually is.
The pattern
Five of six are decisions made before the tool is chosen. That is the argument of this article in one line: the purchase is downstream of a staffing question that most teams never ask.
The one that costs the most
Mistake one, because it is invisible until the renewal. Nobody notices the quarter where the findings were not acted on; the dashboard kept updating and the number kept being reported.
The tell is simple. If you cannot name three things you changed this quarter because of what the tool showed you, the gap is open and the subscription is not the thing to fix.
FAQs
What is an LLM SEO tool?
Software that runs a fixed set of prompts against AI assistants on a schedule and reports whether your brand was named and whether your URLs were cited. Also sold as an AI visibility tracker or AI search monitor. It measures; it does not produce the pages or fixes the measurement calls for.
Why do most teams see no results after buying one?
Because nobody was assigned to act on the findings. Measurement costs a subscription and acting on it costs 25 to 70 hours a month. Teams budget the first number and assume the second, and the work competes with objectives people are actually measured on.
How much does acting on the findings cost?
Roughly 25 to 70 hours monthly depending on how much you intend to fix, dominated by writing pages against gaps. At a loaded rate that is several thousand a month, which puts the subscription at 3% to 15% of the real cost.
Should we hire, use an agency, or buy a managed service?
In-house if you can commit 25 or more hours a month and want to keep the capability. An agency if content volume is the constraint and you have a strategist. A managed service if you have neither and will not hire. Engineering fixes ship through your team in all three.
What can a tool never tell us?
What your product's real constraints are, which claims about it are accurate, and which topics genuinely deserve a page. It also cannot deploy anything to your site. Those stay internal regardless of what you buy.
What should we check before buying anything?
Run twenty prompts manually across three assistants, check server logs for blocked AI crawlers, audit your entity details for consistency, and read your top ten pages' opening paragraphs. Four checks, a day's work, and they usually find the problem.
Is it worth buying if we only publish occasionally?
Probably not. Below about fifty prompts, manual checking monthly costs nothing and works. Buy when manual checking becomes impractical or when you are managing several brands, not before.
How do we know the tool is paying for itself?
Name three things you changed this quarter because of what it showed you. If you cannot, the gap is open. That question is a better renewal test than any movement in the headline rate.
Deciding before you buy
Answer one question first: if a tool told you tomorrow that you lose forty prompts because no page on your site answers them, who writes those pages and when?
If you can name the person and the hours, buy the tool and get started. If you cannot, the subscription will produce a chart and the quarter will pass.
Do these four things this week
Twenty prompts, manually, five runs each across three assistants. Two hours, and you have a baseline.
Server logs by user agent. One hour, and it rules in or out the most common cause of invisibility.
Entity consistency audit across your site, profiles and directories. A few hours of unglamorous and high-return work.
Read your top ten openers. Free, and usually uncomfortable.
Three of those four typically surface the actual problem, and none needs software.
Then pick your route
You have | Route |
|---|---|
25-plus hours a month, committed | Tool plus in-house |
A strategist, no writers | Tool plus agency |
Neither, and no plans to hire | Managed service |
No answer yet | Buy nothing. Build capacity first |
If production is the constraint
If the four checks show your gap is pages you cannot get written, that is a production problem rather than a measurement one. Blazly SEO covers research, planning, drafting and direct publishing from about $39 a month.
It does not track prompts, so it sits alongside measurement rather than replacing it, and the factual review of what it drafts stays with you.
Summary
An LLM SEO tool measures whether AI assistants name and cite you. It does not write the pages, fix the crawler block, correct the directory listing or ship the rendering change, and those add up to 25 to 70 hours a month that almost nobody budgets. Every finding routes to one of four owners across content, engineering and marketing ops, only one of whom usually reports to the buyer. So decide the route before the tool: in-house with committed hours, an agency for content volume, or a managed service if you have neither. Run the four free checks first, because they usually find the problem without a subscription. And judge any tool you do buy on whether you can name three things you changed because of it.