IT Master

SEO automation: what to hand to machines, and what to keep

Topic discovery, research, drafting, validation, publishing, indexing, refresh and measurement can now run without a person. Strategy, brand, link relationships and judgement cannot. Here is where the line sits, and why.

Published 2026-09-02, reviewed 2026-09-04, by Rodd Azad, editor-in-chief. How we check facts.

Eight jobs a machine can now run end to end

Most of the SEO production line is repeatable work with a checkable output, and that is exactly what software is good at. In 2026 these eight jobs can run without a person touching them:

  • Topic discovery. Pull queries with real search demand from DataForSEO and your own Search Console.
  • Research. Read the pages that already rank, your first-party data, and the forum threads where buyers ask the question in their own words.
  • Drafting. Write from that dossier under a fixed brand voice.
  • Validation. Fact-check, score novelty against competitors, critique E-E-A-T, and test for AI tells.
  • Publishing. Push to the CMS with JSON-LD schema, internal links, a hero image and an FAQ block.
  • Indexing. Ping search engines through IndexNow when a page goes live.
  • Refresh. Spot a page whose clicks are decaying and rewrite the parts that went stale.
  • Measurement. Read Search Console nightly and record whether GPT or Perplexity cite the page.

Each job has a defined input and a testable output, which is the condition for automating anything. How it works sets out the pipeline IT Master runs from discovery to publish, and the nightly loop behind refresh and measurement.

In one sentence

SEO automation is the practice of running the repeatable stages of search content production — topic discovery, research, drafting, validation, publishing, indexing, refresh and measurement — as software, while a person keeps strategy, brand and editorial judgement.

Four jobs that stay with a person

Automation is honest only when it names what it cannot do. Four things still need a human, and no SEO automation software changes that.

Strategy. Which markets to enter, which product lines deserve a content cluster, and what to leave alone are commercial decisions. A tool can surface demand; it cannot decide whether the demand is worth your reputation.

Brand. Voice rules, the claims you will never make, the words your customers use and the ones they do not. Software enforces these once they exist. Somebody has to write them first.

Link relationships. Real editorial links come from people who know you: suppliers, trade bodies, customers, journalists. A script that emails strangers is not a relationship, and search engines have spent a decade learning to discount it.

Judgement. Whether a page is true, on-brand and worth publishing under your name is a call an editor makes. IT Master keeps that role explicit rather than pretending it away: the engine has a named Editor-in-Chief, Rodd Azad, and every customer site should name its own, whoever signs off on what goes out under the brand. Plan for it as a standing editorial duty, not a production job.

What Google's spam policy actually says about automation

This is the question every buyer asks and most vendors answer badly, in both directions. Here is what the policy says.

Google's spam policies define scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users", and list among its examples "Using generative AI tools or other similar tools to generate many pages without adding value for users".

Read the qualifier. The violation is the combination of many pages and no added value. Neither automation nor generative AI is prohibited on its own, and the policy names no production method as inherently disallowed. Equally, nobody gets a pass for using a good model: three hundred fluent pages that restate the top ten results are precisely the thing the policy describes.

So the compliance question is not "is this AI-written?" It is:

  • Does each page add something a reader could not get from the pages already ranking? That is what a novelty check measures, and why we run one against the actual competitor pages rather than against a style guide.
  • Is anything on the page checkable? A page of unverifiable assertion adds no value even when it is true.
  • Would you put a named person's name on it? Ours carries one, which is a constraint before it is a marketing line.

Automation makes the answer to those questions consistent across three hundred pages, in either direction. That is the whole risk and the whole opportunity.

In one sentence

Google's spam policy does not prohibit automation or AI writing. It prohibits scaled content abuse, which it defines as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users". The test is value to a reader, not who or what typed it.

The failure mode: generic content at scale

Automation fails in one predictable way. Give a single model a keyword and a word count, run it three hundred times a month, and you get three hundred pages that say what the top ten results already say, in the same rhythm, with the same hedges. Every page is plausible. None is worth ranking.

The damage spreads. Search engines and answer engines judge a site by more than the page in front of them, so a generic cluster can drag on the good pages beside it. Readers bounce, click-through falls, and the refresh loop dutifully rewrites pages that were never the problem.

The cause is structural, not a prompt bug. One model writing and grading its own work approves its own habits, because it cannot see them. It also repeats whatever it read during research, because that is all it has. Volume did not create the problem; volume made it visible. Any SEO automation platform that produces text has to answer this, and naming a favourite model is not an answer.

How cross-vendor checks prevent it

IT Master separates the writer from the judges by vendor. Claude models write. Gemini and GPT models check. The model that writes is never the model that checks, so the writer's blind spots are not the grader's.

Every draft faces four checks before it can publish:

  • Fact-check against the research dossier and the customer's own data.
  • Novelty scored against the competitor pages that already rank; a draft that restates them fails.
  • E-E-A-T critique of whether the page shows experience and evidence, not just fluent prose.
  • AI-tell detection for the rhythms and hedges that mark machine writing.

A failing draft goes back to the writer with specific fixes, up to three times. Only a draft that clears every check is published.

The bar is real. Across 186 Standard-tier runs, 21% of drafts cleared every check on the first attempt with no revision, and 51% got there within the three revisions allowed; the rest never published. Pro-tier judges are stricter: 25% cleared every check across a small sample of 16 runs.

Revising or rejecting that many drafts only works because drafting is cheap. Median generation cost per article: Lite $0.41, Standard $0.85, Pro $1.45 (our cost, not the customer price).

In one sentence

Cross-model validation means the model that writes is never the model that checks: the judges come from a different vendor, so the writer's blind spots are not the grader's.

First-party data is what makes an automated page worth ranking

Cross-vendor checks stop bad pages. They do not, on their own, create pages a competitor cannot copy. That comes from what the writer is given to work with.

Where a customer has product data, datasheets, support Q&A or install notes, IT Master grounds the article in them. A CCTV retailer's guide to choosing a PoE switch can draw on the switch specifications and the questions buyers asked before purchase, rather than a paraphrase of the same three review sites everyone else read. Live tenants publishing this way include UK CCTV trade and retail sites.

The effect on automation is direct. The novelty check has something to measure against, the fact-check has a source of truth, and the finished page carries details that are not sitting on ten other sites. Generic content at scale is what you get when the input is generic; change the input and the output changes with it.

Four automation failures we shipped, and what each one cost

Automation does not fail loudly. It fails by reporting success, which is why an automated pipeline needs measurement of outcomes rather than of steps. These four are ours, with the numbers.

  • A robots.txt rule that reached no crawler. We added rules to stop crawlers walking a combinatorially large faceted URL space, and put them in the wildcard "User-Agent: *" group. A crawler obeys only the most specific group that matches it and ignores every other group, and each site carried thirteen named crawler groups — Googlebot, Bingbot, GPTBot, PerplexityBot, ClaudeBot and the rest — so none of them ever saw the new rules. Measured over 24 hours on one site: 382 of Googlebot's 1,437 requests, 26.6%, went to faceted category URLs across 250 distinct permutations, averaging 3.43 seconds each, against 12 requests to the bare category pages they all canonicalise to. The canonical tags were correct throughout. A canonical does not stop a crawl; Google has to fetch the page to read it. Verify a robots rule by fetching /robots.txt and reading the group for the crawler you care about, never the wildcard group.
  • An indexing channel that accepted everything and recorded nothing. Google's Indexing API answered HTTP 200 for every URL we submitted, with no latestUpdate in the body. Querying the metadata endpoint for those URLs across five sites returned 404 NOT_FOUND on every one — owned, accepted, never recorded. A property we do not own returned 403, so the endpoint does distinguish ownership. Google scopes that API to JobPosting and BroadcastEvent pages. Our submission counters, timestamps and success logs had described work that never happened. A 2xx is not an answer; read the body. IndexNow, a fresh sitemap lastmod and internal links are what actually move discovery.
  • Counting publication as visibility. On 25 August 2026 the share of published articles Google had indexed, across four sites on the same pipeline, was 25 of 27, 8 of 13, 14 of 28 and 12 of 25. Eleven of the worst site's thirteen misses were "URL is unknown to Google" with an empty last-crawl date — never discovered. Sitemap, robots, status code, canonical and pagination all checked out. Publishing volume is an input; indexed share is the output, and it is one SQL query.
  • A scored check used as a gate. Our fact-check reported a truncated link in a draft verbatim and the article published anyway, because the check scored 84 against a passing bar of 78. A check that contributes to an average cannot enforce a floor. Anything that must never ship needs deterministic code — we now run a pre-publish link guard, and the back-catalogue sweep it triggered found 46 dead links across 16 of 92 live articles.

The common lesson is the one worth stealing: automate the check of the outcome, not just the step. Every one of these systems logged success.

In one sentence

The recurring shape of an automation bug: a guard that is installed but inert reads exactly like a guard that is working. Every one of these reported success while doing nothing.

IT Master as the automation layer in your stack

IT Master is not a dashboard of suggestions. It is the layer that runs the eight automatable jobs and hands the result to the site you already have.

Delivery matches the platform:

  • WordPress: install the plugin, paste one key, and articles arrive as posts with Yoast or Rank Math fields filled.
  • Next.js: the @itmaster/sdk package.
  • Anything else: a REST pull feed, or a signed push webhook to any endpoint you control.
  • Shopify and Webflow: set up by us on request. Wix and Squarespace: a copy-paste queue.

Four quality tiers, Lite, Standard, Pro and Premium, set how strict the judges are. You pay per published article from a prepaid balance, with no subscription, and a draft that fails the checks is not charged.

After publishing, IndexNow pings the search engines, the nightly loop watches Search Console for decay and refreshes what has slipped, and the engine records whether GPT or Perplexity cite the page. Full detail is on how it works.

When not to automate

There are situations where the honest recommendation is to publish less, by hand.

  • No first-party data and no search demand. The engine grounds articles in what you know and what people search for. With neither, it has nothing to work with, and what comes out is the average of the pages already ranking.
  • A site with an indexing problem. If Google is not crawling what you publish now, publishing four times as much makes the crawl budget problem worse, not better. Fix discovery first.
  • Regulated claims. Medical, legal, financial and safety copy needs a qualified human signing it off, per page. Machines can draft; they cannot carry the liability.
  • A brand whose value is one person's voice. If readers come for a named writer, a pipeline is the wrong tool, whatever the pass rate.
  • A site that needs links, PR or a rebrand more than it needs pages. Do those first. Content compounds on top of authority; it does not substitute for it.
  • Anywhere you cannot measure the result. Without Search Console, analytics and a citation probe, an automated pipeline is publishing into the dark, and the refresh loop has nothing to steer by.

One more, on cadence rather than fitness: do not dump. Volume is safe when it arrives on a steady schedule alongside genuine measurement, and conspicuous when three hundred pages appear in a week. Our own publishing is drip-fed for that reason.

In one sentence

Automation multiplies whatever it is pointed at. If the underlying content decision is wrong, automating it produces the wrong thing faster and at a volume that is harder to withdraw.

How to evaluate SEO automation tools before you buy one

Whatever you choose, run the same questions against it. They separate SEO automation software that removes work from software that moves the work onto you.

  • Does it publish, or hand you drafts? A tool that stops at a draft has automated the cheapest stage and left you the rest.
  • Who checks the writer? Ask which model writes, which model grades, and whether they come from different vendors. One vendor is one set of blind spots.
  • What is it grounded in? If the answer is the top ten results, the output will be the average of them.
  • Does it measure and refresh, or only produce? Search Console and AI citations are the feedback loop; without them the tool cannot learn what worked.
  • What do you pay for a failure? If a rejected draft costs the same as a published page, the incentive is to reject nothing.

You can start with our free tools before speaking to anyone: run your site through the site check, and see whether AI assistants already cite you with the AI visibility checker.

Sources

Primary sources for the policy and protocol claims on this page, all read on 4 September 2026.

Questions people ask

Can SEO be fully automated?

No, and anyone who says otherwise is selling the wrong thing. The production stages can be automated end to end: finding topics with real demand, researching them, drafting, validating, publishing, pinging the index, refreshing decayed pages and measuring results. The decisions around them cannot. Which markets to enter, what your brand will and will not say, which relationships earn you links, and whether a page deserves your name are human calls. A well-built system takes the first list off your desk and leaves you the second, which is a standing editorial duty rather than a daily production job.

What does SEO automation software actually do?

It depends heavily on the product, which is why the phrase covers so much. At one end are audit tools that crawl a site and produce a list of things for you to fix. In the middle are writing assistants that produce a draft you then edit, upload and format. At the other end are engines like IT Master that run the whole loop: discover a topic from Search Console and keyword data, research it, write, check the draft with models from a different vendor, publish to your CMS with schema and internal links, ping IndexNow, then measure and refresh. When comparing tools, ask where in that range each one stops.

Is automated content a risk for rankings?

Generic content is the risk, not automation. A page that restates what already ranks, whoever produced it, gives a search engine no reason to show it and a reader no reason to stay. Nobody can promise how any page will be treated, so the honest approach is to remove the causes of generic output: ground the writing in data competitors do not have, and have models from a different vendor reject drafts that lack novelty or evidence, or that carry the telltale patterns of machine prose. That bar is measurable. Across 186 Standard-tier runs, 21% of drafts cleared every check on the first attempt with no revision, and 51% got there within the three revisions allowed; the rest never published.

How much does SEO automation cost with IT Master?

You pay per published article from a prepaid balance. There is no subscription, and a draft that fails the validation checks is not charged, so you pay for pages that reached your site rather than for attempts. Four tiers, Lite, Standard, Pro and Premium, let you match the strictness of the checks to how much a page matters, and each tier carries its own per-article rate. Current rates are on the pricing page, alongside what a typical month costs at the cadence you want to publish.

What do I still have to do myself?

Three things, and they are the ones worth your time. First, set the strategy: which topics and product lines you want to be known for, and which you do not. Second, define the brand: the voice, the terms you use, and the claims you never make, so the writer and the judges have something to enforce. Third, act as editor-in-chief: review what ships and steer what comes next. Link relationships also stay human, because real editorial links come from people who know your business. Everything else, from research to indexing to refresh, is the engine's job.

Will Google penalise AI-written content?

Not for being AI-written. Google's spam policies target scaled content abuse, defined as many pages generated primarily to manipulate rankings rather than help users, and list generative AI used to produce many pages without adding value as one example of it. The qualifier is the whole rule: the violation is volume without value, not authorship. In practice that makes the useful question whether each page adds something the pages already ranking do not, which is exactly what a novelty check against those competitors measures.

What is the smallest useful amount of SEO automation?

Three jobs, in this order. First, automated measurement: Search Console data pulled on a schedule and one query that reports what share of your published pages is actually indexed. Second, automated discovery submission through IndexNow when a page changes. Third, a decay report that lists pages whose impressions have fallen for a month. None of those writes anything, all three are cheap, and together they tell you whether writing more is your problem at all. Automate production after you can measure it.

How do I tell whether an automation is actually working?

Check the outcome, never the step. Automated pipelines fail by reporting success: a submission endpoint returning 200 while recording nothing, a robots rule sitting in a group no crawler reads, a publish counter rising while indexed share stays flat. For every automated step, name the observable outcome it is supposed to produce and measure that separately, from a different source where you can. If a guard has never once refused anything, treat it as broken until you have proven otherwise.

Read next

See what search engines and AI assistants find on your site

Free, and no account. Type your address and the check tells you what is missing and what we would write first.

Check my siteSee what an article costs
Run the free site checkPricing