By KYN AI Advisory Team — AI implementation specialists, Singapore
A weekly Search Console review takes five query lists: queries near the top of page two, high-impression low-click queries, new queries, lost queries and queries landing on the wrong page. An agent pulls and sorts them from the Search Console API. A person decides which three to act on. The agent prepares the list. It does not edit pages or publish anything.
What data does the agent read?
Search Console reports clicks, impressions, click-through rate and average position, by query, page, country and device. The Search Analytics API returns the same data as the Performance report, so an agent can pull it on a schedule without anyone opening the interface.
Give the agent read-only access to one property. It needs no write rights. In a custom build, the agent runs under its own restricted account, the credentials sit in your cloud account, and every pull is logged.
Two cautions on the data. Position is an average over the period, so a query that shows position 11 may have appeared at 4 for some searches and 30 for others. And Search Console hides some rare queries for privacy, so the totals by query will not add up to the totals by page. Read both views.
Which five lists should it produce each week?
Compare the last 7 days with the 7 days before, or the last 28 with the previous 28 for low-traffic sites. Short windows on a small site are noisy.
- Near page one. Queries with average position between 8 and 20 and meaningful impressions. These are the cheapest wins: a better answer paragraph, a table or an internal link can move them.
- High impressions, low clicks. Queries where you appear often but few people click. The title or meta description is probably failing to state the answer. Rewrite both.
- New queries. Queries that did not appear before. They show how buyers phrase the question. Add them to your headings and FAQs, or write a new article if none of your pages fits.
- Lost queries. Queries that dropped in clicks or position by a set margin. Check whether the page changed, a competitor page appeared, or the query itself faded.
- Wrong page. Queries where the page receiving impressions is not the page you would choose. Often two of your pages compete. Pick one and link the other to it.
Each list is a table: query, page, clicks, impressions, CTR, position, change. The agent writes one line per row saying what changed. It does not guess the cause.
What does a person do with the lists?
Limit the action to three items a week. More than that does not get done.
| List | Typical action | Time | |---|---|---| | Near page one | Tighten the answer paragraph, add a table or FAQ, add two internal links | 20 to 40 minutes | | High impressions, low clicks | Rewrite title and meta to state the answer | 10 minutes | | New queries | Add an H2 or FAQ using the query wording | 15 minutes | | Lost queries | Compare the page with the current top results, restore what was cut | 30 minutes | | Wrong page | Choose a primary page, link the other to it | 15 minutes |
Edit the live page in place. Do not delete or redirect a page to fix a ranking problem, because the page's history is what you are trying to keep. See why you should never delete a blog article.
After a change, wait. Rankings take weeks to settle. Log the date and the change in a sheet so you can compare against the same query four weeks later.
What should the agent never do?
- Edit a live page. It can propose a new title. A person approves it.
- Claim a cause. "Position fell because of the update" is a guess. The agent states what changed in the numbers and stops.
- Act on tiny samples. Ignore rows under a minimum impression count, set at the start.
- Mix properties. One property per report, with the country filter stated.
How do AI answers change the review?
Search Console shows clicks from Google Search results. It does not show whether an AI assistant cited you. Pair the weekly review with a monthly citation check. See the competitor citation audit and how to appear in AI search answers.
What does the weekly output look like?
One page, same layout every week:
- the date range compared;
- totals: clicks, impressions, average CTR, with change;
- the five lists, top 10 rows each;
- three suggested actions, ranked by impressions affected;
- last week's actions and their effect.
Sent to a named person by email or WhatsApp. The agent can post it. The reply, "approve 1 and 3", is what starts any change.
What does one weekly report look like with numbers?
The figures below are illustrative. They show how to read the lists, not what your site will show.
| Query | Position | Impressions | Clicks | CTR | Action | |---|---|---|---|---|---| | bank reconciliation singapore | 11.4 | 1,800 | 22 | 1.2% | Tighten the answer paragraph, add a table | | gst filing deadline | 6.2 | 2,400 | 31 | 1.3% | Rewrite title and meta to state the date | | payroll cpf 2026 | 14.8 | 640 | 5 | 0.8% | Add an FAQ using the query wording |
The first row sits on page two with real impressions, so it is a near-page-one candidate. The second row ranks well but earns few clicks, so the snippet is the problem, not the ranking. The third row has low volume, so it goes below the others in the ranked list. The agent orders the rows by impressions affected and the person picks three.
How do you set the thresholds for each list?
Write the thresholds down at the start so the lists do not change shape from week to week.
- Near page one. Position 8 to 20, and a minimum impression count such as 100 over the window.
- Low clicks. Position under 10 and a CTR below the median for your pages.
- New query. No impressions in the previous 28 days.
- Lost query. Clicks fell by more than half against the prior window, with a minimum prior click count.
- Wrong page. The page with the most impressions for the query is not the page assigned to it in your content plan.
The numbers are yours to choose. A site with 200 impressions a week needs lower floors than one with 20,000. Review the thresholds each quarter.
How do you record what changed?
Keep a change log with one row per edit: date, page, query, the change and the baseline numbers. Four weeks later the agent adds the new numbers beside the baseline. Without the log, you cannot tell whether an edit helped or the query moved for another reason.
Note changes you did not make. A competitor publishing a new page or Google changing a results layout shifts numbers with no edit on your side. The agent records the date and leaves the explanation to you.
Where KYN fits
KYN builds the weekly review as part of marketing systems: a scheduled pull from Search Console, the five lists, and a short message to a named person. It runs in a cloud account in your name, the agent holds a restricted read-only account, and every action is logged with the actor. Changes to live pages wait for a person. KYN's own AI search check costs about S$0.05 per check, which gives a sense of the unit cost of this kind of narrow job.
The first 14 days are a review of your stack and interviews with your team. A full build ships in under 90 days. See marketing.
Frequently asked questions
How much data do I need before this is useful?
Enough that a week has meaningful impressions for your top queries. For a small site, use a 28-day window instead of 7 days, and ignore rows with very few impressions.
Can the agent rewrite my titles automatically?
It can draft them. It should not publish them. A title is the first thing a buyer reads, and a person should approve the wording.
Is the Search Console API free?
Google provides the API for properties you own, with usage limits described in its API documentation. Check the current limits there. A weekly pull for one property is a small number of requests.
Checked 30 October 2026.
Curious whether Marketing agents fits your business? Talk to KYN on WhatsApp — no forms, just a conversation.
