How to tell if your agent-drafted blog posts are actually working, 30 and 90 days in
You published the post. The draft came out of the agent loop clean, a human approved it, and it went live. Now what?
Most teams either check obsessively for a week and give up, or they never look at the post again. Both are wrong. Agent-drafted content earns the same evaluation window as any other post, and the honest answer usually shows up around 30 days for early signals and 90 days for real judgment.
Here is the framework I use, and the three fixes that most often bring a stuck post back to life.
Why 30 and 90 days, not 7
Google needs time to crawl, index, evaluate, and then re-evaluate. Fresh posts get a small honeymoon in the index, then drop, then settle. If you judge a post at day 7, you are grading it during that noisy settling period.
Thirty days is the first honest check-in. By then Search Console has enough impression data to show whether Google understands what the post is about. Ninety days is where you make the call: patience, tweak, or rewrite.
AI answer engines are less predictable. Some posts start showing up as citations in Perplexity or ChatGPT search within days, others never do, and the pattern is still moving month to month. Treat citation pickup as a signal to watch, not a metric to forecast.
The four metrics that matter after publishing
Skip the vanity dashboards. Four numbers actually tell you what is happening.
1. Impressions in Search Console
Impressions are the honest first signal. They tell you how often your post appeared in a search result, whether or not anyone clicked.
At 30 days, look for a rising line. Not a huge one. A post targeting a small cluster keyword might sit at 50 to 300 impressions per day and still be healthy. A post targeting a bigger term should be climbing, not flat.
Flat impressions after 30 days mean one of two things: Google has not decided what the post is about, or it has decided and the answer is "not this query." Both are fixable, but you need to know which.
2. Position drift
Average position is a bad standalone number, but the change in position over time is a great one.
Open the Performance report, filter to the post's URL, and compare weeks. A healthy post moves from position 40 to 25 to 18 over a month or two. A stuck post lands at position 30 and never moves. A dying post starts at 12 and drifts to 22.
Position drift also tells you which query Google is actually ranking you for. Sometimes a post you wrote for "content operations checklist" starts ranking for "content ops template." That is a signal, not a failure. Rewrite the H1 and intro to match what Google already thinks you are about.
3. Assisted conversions
Direct conversions from a blog post are rare and usually a coincidence. Assisted conversions are the real business metric.
Set up a simple attribution view in GA4: last non-direct click, then look at pages that appeared anywhere in the conversion path. A post that never converts directly but appears in 40 percent of paths for a specific product page is doing its job. A post with high traffic and zero path involvement is entertainment, not marketing.
For B2B and SaaS teams, pair this with a self-reported "how did you hear about us" field on your signup form. It is unscientific, and it is the closest thing you get to ground truth.
4. Citation pickup in AI answers
This is the metric most teams still ignore, and it is the one that will matter most over the next year.
Test your post the way a reader would. Open Perplexity, ChatGPT with search on, and Google AI Overviews. Ask the two or three questions the post is meant to answer. Are you cited? Is a competitor cited from a weaker page? Is the AI paraphrasing your exact framing?
Track this monthly in a plain spreadsheet. Post URL, target question, cited yes/no, citation position. That is enough. Over a quarter you will see which posts are becoming reference material for AI answers and which are invisible to them.
Separating a post that needs patience from one that needs a rewrite
At the 30-day check, most posts fall into three buckets.
Growing. Impressions rising, position improving, some clicks. Do nothing. Come back at 90 days.
Flat with signal. Some impressions, position holding around 20 to 40, occasional click. This is the "patience or nudge" bucket. If the topic is competitive and the post is under a month old, wait. If it has been eight weeks with no movement, intervene.
Flat with no signal. Under 20 impressions per day, no position anchor, no clicks. Google does not know what this post is for. Rewriting the intro and title usually helps more than adding words.
The trap is treating bucket two like bucket three. A post that is climbing slowly does not need surgery. Every rewrite resets the clock in Google's index and can hurt more than help. Do not touch a post that is still moving in the right direction, even slowly.
The three interventions that most often move a stuck post
When a post genuinely needs help, these three fixes account for maybe 80 percent of the recoveries I have seen.
1. Fix the intent match
Search intent is the most common mismatch, and it is the hardest one to see in your own writing.
Take the top three ranking URLs for your target query. Read them. Are they how-to guides? Comparisons? Definitions? Tool roundups? If your post is a 1,500-word narrative essay and the top three are structured comparison tables, you are answering the wrong shape of question, no matter how good the prose is.
This is where the research step in the agent loop earns its keep. A good research agent should show you the intent pattern before drafting, not after. If your post missed the intent, do not add more sections. Restructure the top of the post to match what the query actually wants, then keep the depth below.
2. Add internal links from stronger pages
Every site has a small number of pages that carry most of its authority. Usually the homepage, a pricing or product page, and two or three cornerstone posts. A stuck post almost always has zero internal links from these pages.
Audit your top five most-linked pages. Add a contextual link to the stuck post from at least two of them. Not a footer link. Not a "related posts" widget. A real, in-body link from a sentence that a human would actually click.
This is the single fastest lever for a post that Google has indexed but not promoted. Give it a few weeks to re-crawl.
3. Rewrite the opening
Readers, and increasingly AI answer engines, decide in the first three sentences whether the post is worth quoting.
Look at your first paragraph. Does it restate the question, hedge, and promise value later? That is the standard AI-drafted opening pattern, and it kills both dwell time and citation rate. Rewrite it as a direct answer to the query, in one or two sentences, then earn the rest of the reader's attention. The same craft rules that keep a team blog readable apply here: lead with the answer, cut the wind-up, respect the reader's time.
An editor agent is genuinely useful here. Ask it to rewrite the opening as "the answer, in two sentences, then the reason it is not that simple." That single change often lifts a post out of the flat-with-signal bucket.
What to do at 90 days
At the 90-day mark, make one call per post.
Keep the growing ones. Do not touch them. Add them to your internal linking rotation so new posts can borrow their authority.
Rewrite the flat ones that never got signal. Treat the rewrite as a new draft, not a patch. Run it back through the research and writer steps, get a human approval, and republish under the same URL.
Retire the ones that ranked briefly and then dropped for a substantive reason, such as a product change, an outdated framework, or a topic the business no longer sells against. A shrinking archive is a healthier archive. This is the same judgment as knowing when a cluster is done, applied one post at a time.
Measurement is part of the loop, not an afterthought
The mistake most small teams make is treating publishing as the end of the workflow. It is the middle. A post budget only pays off if the posts that miss get fixed and the ones that land get amplified.
Build the 30 and 90-day check into your calendar the same way you built the drafting step. Ten minutes per post is enough. Teams that do this consistently end up with a compounding archive. Teams that do not end up with a graveyard of drafts that a human once approved and then everyone forgot about.
Measurement is part of the loop, and the loop is what RankMill runs. Research agent finds the opportunity, writer drafts it, editor takes revisions, a human approves, and then the 30 and 90-day check tells you which posts to feed and which to retire. Nothing ships without your say-so, and nothing lives on unexamined. Made by agents. Approved by humans.