The agency replacement
Most people looking for an AI SEO agency want the work, not the agency.
Searchlings is software that does the production work an AI SEO agency does: keyword research, technical audits, content drafting, verification and publishing. It runs every night, logs every action with a timestamp, and sends you one report in the morning.
The work
What an AI SEO agency actually does.
An AI SEO agency sells the same list of jobs a traditional SEO agency sells, with software doing more of the production. Strip the branding off any of them and the deliverables are close to identical: find the terms your customers search, audit the site for the technical faults that stop those pages being found, rewrite the pages that are underperforming, publish new pages for the questions you do not answer yet, earn or ask for links, and report on what moved.
Most of that work is production. It is repetitive, it follows rules, and it has to happen continuously rather than in bursts. A small number of the jobs on that list are genuinely judgement: deciding what your business should be known for, choosing which market to fight for, knowing which claim your industry regulator will object to.
The reason this page exists is that the two halves are usually sold as one thing, at one price, on one retainer. Splitting them changes what you pay for and what you can check.
The jobs, and which half they fall into
| The job | Production or judgement |
|---|---|
| Crawling the site and finding technical faults | Production |
| Keyword and opportunity research | Production |
| Writing and rewriting page copy | Production |
| Checking work against an editorial standard | Production |
| Publishing, and submitting for indexing | Production |
| Tracking rankings and AI citations | Production |
| Reporting what changed | Production |
| Deciding what the business should be known for | Judgement |
| Positioning, brand and messaging | Judgement |
| Digital PR and link outreach | Judgement |
| Planning and supervising a site migration | Judgement |
| Knowing what your regulator will object to | Judgement |
Searchlings does the first seven. It does not do the last five, and this page says so again further down, because a supplier who is vague about the boundary is a supplier you will argue with later.
The words
AI SEO, AEO and GEO: what the words mean.
The category has acquired four names in about two years, and the vocabulary is doing a lot of work in sales conversations. It is worth knowing which distinctions are real.
AI SEO is used two different ways, and the ambiguity is not accidental. It can mean using AI to do SEO work, or optimising your site so AI systems mention you. Most agencies selling AI SEO mean the second. Searchlings does both, and separates them deliberately, because they are different jobs with different evidence.
AEO, answer engine optimisation, is the work of being named in the answer an assistant gives rather than ranked in the list of links beneath it. GEO, generative engine optimisation, came from a 2023 research paper and is now used more or less interchangeably with AEO. Wikipedia's own article on the subject notes that the academic literature has not settled on a definition separating the two.
Google's position is blunter. Its search documentation, updated in July 2026, says that from Google Search's perspective, optimising for generative AI search is optimising for the search experience, and thus still SEO. We think that is broadly right, which is why no part of this product is named after an acronym. If the vocabulary turns over again, the work does not change.
Longer versions of both arguments are in the Answer Engine Optimization article and the Generative Engine Optimization article, with the definitions in the glossary.
The retainer
What you are actually paying for.
This is not a claim that agencies are useless. It is a claim about where a retainer goes.
A typical retainer
- An account manager, and the meetings that justify one
- A monthly report summarising work you did not see happen
- Content written to a word count, checked by whoever had time
- An audit when someone gets to it
- A renewal conversation
Searchlings
- No account manager and no meetings
- A log of every action, timestamped, that you can read at any time
- Content checked against a fixed standard, or rejected with numbered reasons
- A crawl every night, whether or not anyone is watching
- Cancel from your account, no phone call
The awkward part of an agency retainer is not the price. It is that the unit you buy is time, and time is the one input you cannot verify. You are told six hours went into your account last month. There is no artefact that proves it, and no artefact that disproves it either, which is why the argument is always about the report rather than about the work.
Software has the opposite property. It cannot pretend to have done a crawl. Either there is a crawl record with a timestamp and a page count, or there is not.
The number
What this costs, and what an agency costs.
Almost nobody in this market publishes a price. Of the agency service pages ranking for this term, none of the ones we checked put a figure on the page. The comparison pages do, usually by quoting somebody else's rate card.
Ours is on the pricing page and it is on this one. Starter is $299 a month and Growth is $699 a month, in Australian dollars, including GST. There is no setup fee, no minimum term and no exit fee.
For an outside reference point on the other side: Searchbloom, a United States agency that publishes bands rather than hiding them, quotes roughly US$1,000 to US$5,000 a month for small and local businesses, and US$5,000 to US$15,000 for mid-market. Those are US figures in US dollars and Australian retainers vary, so treat them as an indication of the shape of the market rather than a like-for-like conversion. We are not going to invent an Australian average we have not measured.
The honest way to read that gap is not that agencies are overcharging. It is that a retainer buys people, and people are expensive and finite. A retainer at the bottom of that band buys a few hours a month of someone's attention. The same money here buys work that runs nightly, whether or not anyone is thinking about your account.
Where an agency is genuinely worth its retainer is the judgement column in the table above. If what you need is a strategist, buy a strategist. If what you need is for the production work to actually happen every week, that is a different purchase.
The team
Five workers, and what each one is not allowed to do.
Most suppliers in this category will tell you AI is involved and leave it there. Here is the actual division of labour, including the limits.
- 1
Scout
Finds the opportunities. Watches what people search for, spots the gaps your competitors missed.
- 2
Digger
Keeps your site healthy. Crawls every page, finds what is broken, fixes what it can, flags what it can’t.
- 3
Scribe
Writes the content. Drafts articles and pages in plain language, built around what Scout found.
- 4
Checker
Verifies everything. Nothing publishes until it passes the standard. No exceptions.
- 5
Herald
Publishes and keeps score. Puts passed content live, then tracks where you appear in search results and AI answers.
What none of them do: decide what your business is, invent a fact about you, approve their own work, or edit a page you have marked as protected. Scribe cannot publish, because publishing belongs to Herald and only after Checker has passed the draft. Checker cannot be overridden to let something through. The separation is the point, and it is enforced in the pipeline rather than in a policy document.
The standard
The difference is what gets stopped.
Any tool can draft faster. The question worth asking a supplier, human or software, is what happens to work that is not good enough. Here, it does not publish. Checker runs every draft against the standard and the verdict is passed or rejected, with reasons. No middle verdict, and no way to wave something through because it is Friday.
The standard is a fixed set of checks: word count range, keyword frequency, link accuracy, sentence and comma limits, banned clichés, and repeated-phrase detection. It is identical on every plan. There is no tier where checking is optional, because a plan that sells you weaker checks is selling you a worse product at a discount.
You can read every verdict on your own content. The same checks run against these pages before they deploy, which is why this site has the standard applied to it rather than described by it.
See how Checker worksDraft: “Choosing a commercial cleaning contract”
- Sentence over the comma limit (paragraph 4)
- Brand mention missing
- Source published 2019, outside the freshness window
Control
Nothing goes live without your approval.
This is the question nobody in this market answers, and it is the first one a sensible buyer asks. If software is editing your website, who said it could?
Every proposed change is a changeset: a specific edit to a specific page, with the old version and the new version stored side by side, and a reason attached. A changeset carries a risk tier. Low-risk work can apply automatically if you allow it. Everything else waits in a queue until you approve it, and waits indefinitely if you never do.
You set the tiers per category, so you can let meta descriptions update themselves while requiring approval on anything that touches body copy. Pages you mark as protected are never edited automatically at any tier. For customers in regulated industries the profile forces every category to approval, and that setting cannot be relaxed by the software.
Approving from a phone was a design requirement rather than a nice extra, because approvals happen in the ten minutes someone has between other things. If the queue is only usable at a desk, it does not get read, and a queue that does not get read quietly becomes a queue that gets approved without reading.
Safety
What happens when a change makes things worse.
Some changes will make things worse. Anyone who tells you otherwise is describing a product they have not run for very long.
Every changeset stores a snapshot of the page as it was before the edit. Reverting is restoring that stored version, not reconstructing it from memory or from an export someone took last quarter. That is a small architectural decision that turns a bad afternoon into a button.
Applied changes are then watched for a set window. At the end of it the change is judged against what the page was doing before, and the verdict is written down. When the data is too thin to support a verdict, which so far has been the usual answer for a small site, it says so rather than producing a confident number from nothing. A missing measurement is reported as missing.
Nothing is deleted from the record. Superseded, blocked and reverted changesets stay readable, because a log that quietly loses its failures is not a log.
The objection
Will Google penalise AI-written content?
This is the objection the category avoids, and avoiding it is worse than answering it. At least one well-known agency ranking for this term says on its own page that it will not use AI to write SEO content at all. It is a reasonable position and it deserves a direct reply.
Google's stated position is that it rewards helpful content and acts against content made primarily to manipulate rankings, and that how the content was produced is not itself the test. The risk has never really been authorship. The risk is thin, duplicated, generic or inaccurate pages, which agencies and freelancers have produced in enormous quantities for twenty years without anyone calling it an AI problem.
So the honest version of the argument is not that machines write better than people. It is that the failure mode is known, and you should ask any supplier what specifically stops it. Most cannot answer. Ours is a deterministic set of checks that runs before anything publishes, produces a pass or a reject with numbered reasons, and cannot be waved through. You can read the verdicts.
There is a second reason to care about quality here, which is that the checks are also what makes a page quotable by an assistant. A page that answers its own question in the opening paragraph is easier for a model to lift, and easier for a person to read. Those turn out to be the same edit.
Week one
What happens in the first week.
Nobody in this market tells you what day one looks like, which is odd, because it is the part a buyer can actually verify.
- You send a Scout. Before you pay anything, Scout crawls the site and returns a short report on how findable it is, including whether assistants mention you. If your platform is one we cannot publish to, the report says so at that point rather than after you have paid.
- The first full crawl runs. Digger goes through the site properly and files what it finds as issues, in plain language, sorted by what is worth doing rather than by severity score.
- Research lands. Scout builds the keyword and opportunity picture, including the terms competitors rank for and you do not.
- The first changesets appear in your queue. Real proposed edits to real pages, each with the before and after and a reason. This is the point where you find out whether you agree with its judgement, and it is deliberately early.
- The reports start. One in the morning, short, in plain language, saying what ran and what it needs from you.
What does not happen in the first week is a ranking change. Crawling and drafting are fast. Search results are not, and we do not publish a typical timeline for them, because the honest answer depends on your site, your market and your competitors, and any single number we printed would be a guess wearing a suit.
AI answers
Getting named in AI answers.
When someone asks an assistant a question in your market, your business is either named or it is not. There is no page two, and no partial credit for nearly being mentioned.
Herald asks the questions your customers ask, on a schedule, and records which businesses were named and which pages were cited. It keeps the answer text it came from, so a number on the dashboard can be traced back to the reply that produced it.
Two honest qualifications. This is sampling rather than measurement: assistants do not return the same answer twice, so the trend over weeks is the signal and any single answer is noise. And the useful output is usually not the citation count, it is the list of questions where you never appear, because that is the part you can do something about.
The mechanics, the limits and the sample counts are on the AI citations page.
When to hire
When you should still hire someone.
Being straight about this is cheaper than a refund. Software is the wrong choice when the job is a conversation. Keep a consultant or an agency if you need positioning and brand strategy, digital PR and link outreach, a migration planned and supervised, or someone accountable in a room with your leadership team.
Searchlings covers the production work underneath all of that: the research, the audits, the drafting, the checking, the publishing and the reporting. Running both is a reasonable arrangement, and it means paying a strategist for strategy rather than for production.
It is also the wrong choice if your site has no real content problem and no technical problem, which does happen. Send a Scout first. If the report comes back saying there is not much to fix, that is a useful answer and it costs nothing.
Limits
What we do not promise.
We do not guarantee rankings, and we would treat a guarantee from anyone else as a reason to ask harder questions. Nobody controls the ranking systems, and a supplier who promises a position is either describing a term nobody searches or hoping you will not check.
We do not publish projected results, illustrative traffic charts, or a typical customer outcome. When we have enough real data to say something useful about outcomes, we will publish it with the sample size attached and the method described. Until then the honest position is that we sell the work, not a result.
What we do guarantee is the production: the crawls run on schedule, every draft is checked against the standard, every action is logged with a timestamp, and you can read all of it, including the failures.
Questions
Questions.
Is Searchlings an AI SEO agency?
No. It is software. There is no account manager, no onboarding call and no retainer. You point it at your site and it runs the work an agency would run as production: research, audits, drafting, checking, publishing and reporting.
What does an agency do that Searchlings does not?
Strategy conversations, brand and positioning work, digital PR and link outreach, and anything requiring someone to sit in a meeting with your team. Keeping a consultant for those while Searchlings runs the production work underneath is a sensible arrangement.
How is this different from an agency that says it uses AI?
Most of them use AI to draft faster and still bill for the time. The difference that matters is not who drafts, it is what stops bad work publishing. Every draft here passes a fixed standard or is rejected with numbered reasons, and you can read every verdict.
Will Google penalise AI-written content?
Google penalises content made to manipulate rankings rather than to help people, and it has said that how content is produced is not itself the test. Thin, duplicated, inaccurate work is the risk, whoever or whatever produced it. That is the specific thing the standard is built to stop, which is why every verdict is readable rather than asserted.
Who approves changes before they go live?
You do, for anything above the automatic tier. Every proposed change is a changeset with a risk tier: low-risk edits can apply automatically if you allow it, and everything else waits for you. Pages you mark as protected are never edited automatically, regardless of tier.
What happens if a change makes things worse?
Every change stores a snapshot of what the page said before it was edited, so reverting is restoring a saved version rather than reconstructing one. Applied changes are watched for a set window afterwards, and a change that looks worse on the numbers is flagged rather than left alone.
How long before it does anything?
The audit and the first proposed changes arrive in the first week, because crawling and research are fast. Search results move on their own schedule and we do not publish a typical timeline for that, because we would be inventing it.
Do you guarantee rankings?
No, and you should be cautious of anyone who does. We guarantee the work: the crawls run, the drafts are checked against the standard, the log is complete, and you can read all of it.
Can I use this alongside my existing agency?
Yes. The split that makes sense is the agency keeping strategy, positioning and outreach, with Searchlings running the production work underneath. The log makes it obvious which changes came from where, which tends to help rather than complicate that relationship.
What happens to the work if I leave?
Published content stays on your site and you keep it. There is no rented-content arrangement and nothing is removed when you cancel.
Is there a contract?
No. It is a monthly subscription that you cancel from your own account without a phone call. Annual billing is available at a discount if you prefer it, but it is not required.
Start with a Scout.
Give us a web address. Scout will take a look and send you a short, honest report on how findable you are today, including whether AI assistants mention you. Free, no account needed.
Send a Scout