On-Page SEO, Straight From Search Console Data

Problem

Paid SEO tools charge for something Google already gives away: a per-page breakdown of every search query that’s actually shown your content, how often people clicked it, and where it ranked. That data lives in Search Console’s Performance report the moment a page starts getting impressions. Most people never open it, and even when they do, a spreadsheet of forty queries and their click-through rates doesn’t turn itself into “change this sentence,” that translation step is the part a paid tool, or an SEO specialist, usually gets hired to do.

There’s a second wrinkle before you can even write the prompt: how much of the page can you actually change? On your own site or CMS, you can rewrite the title tag, the meta description, the headers, all of it. On a platform someone else administers, Medium, a company blog you don’t own, a guest post, LinkedIn, you’re often limited to the body copy itself; the metadata is locked or managed elsewhere. Optimizing for the second situation with advice meant for the first wastes effort suggesting changes nobody can actually make.

Pattern

One export, one prompt, in two flavors depending on how much of the page you control.

Get the data

1. Open Search Console and select the property the page belongs to.
2. Go to the Performance report, click ”+ New” → “Page”, paste the exact URL of the page you’re optimizing, and apply the filter. Everything below now describes just that one page, not your whole site.
3. Stay on the Queries tab (it’s the default one after filtering). This is the table that matters: every search query that actually earned an impression for this page, plus its clicks, click-through rate, and average position.
4. Widen the date range to at least the last three months. A single page’s query list on a 7- or 28-day window is often too thin to prioritize against.
5. Click Export → CSV (or “Google Sheets” if you’d rather skim it there first) to download the table.

If the page is new or low-traffic, this table may be short, or a few very-low-volume queries may show blank click or position values, Search Console suppresses some rows for privacy at low volume. That’s normal; work with what’s there. This pattern gets more useful as a page accumulates weeks of real data, not less.

Grab the page’s existing copy

Copy the live text of the page (or the draft, if it isn’t published yet) into a plain text file. If you’re using the full-control flavor below, also grab the current title tag and meta description; your page’s source or your CMS’s SEO fields will show both.

You're helping me optimize this page's copy for SEO. I can only edit the body copy on this
page, not the title tag or meta description, so don't suggest changes to either.

Here's the page's current copy:
[PASTE FULL PAGE COPY]

Here's a Search Console export of every query this exact page gets impressions for
(columns: Query, Clicks, Impressions, CTR, Position):
[PASTE CSV DATA]

1. Group the queries into: (a) high-impression queries with a weak click-through rate,
   (b) "striking distance" queries around position 8-20 with real impressions, and
   (c) queries with real apparent volume that the current copy doesn't address at all.
2. Tell me the 3-5 highest-opportunity queries overall and which group each falls into.
3. Rewrite the body copy to work those queries naturally into the opening paragraph,
   subheadings, and supporting sentences, without changing any of the page's actual
   claims or adding anything that isn't true.
4. If a query doesn't fit naturally anywhere, don't force it in; tell me which ones and why.
5. Match the existing tone and voice exactly. Don't let it drift toward generic SEO phrasing.

Return the full rewritten copy, then a short list of what changed and why.

You're helping me optimize this page for SEO. I have full control over the page: title tag,
meta description, headers, and body copy.

Here's the page's current copy:
[PASTE FULL PAGE COPY]

Current title tag: [PASTE CURRENT TITLE TAG]
Current meta description: [PASTE CURRENT META DESCRIPTION]

Here's a Search Console export of every query this exact page gets impressions for
(columns: Query, Clicks, Impressions, CTR, Position):
[PASTE CSV DATA]

1. Group the queries into: (a) high-impression queries with a weak click-through rate,
   (b) "striking distance" queries around position 8-20 with real impressions, and
   (c) queries with real apparent volume that the current copy doesn't address at all.
2. Tell me the 3-5 highest-opportunity queries overall and which group each falls into.
3. Write a new title tag (aim for 50-60 characters) and meta description (aim for
   150-160 characters) that lead with the single highest-opportunity query, honestly
   describe what the page delivers, and read like something a person would click, not a
   keyword string stitched together.
4. Suggest specific header changes only where a heading can pick up an underused query
   without becoming awkward; don't force every heading to change.
5. Rewrite the body copy to work the remaining high-opportunity queries in naturally,
   without changing any of the page's actual claims.
6. If a query doesn't fit naturally anywhere, tell me which ones and why instead of
   forcing it in.

Return: the new title tag, the new meta description, any header changes, and the full
rewritten body copy.

Before you rewrite anything: here's how this page should sound. [Paste 5-10 sentences of
copy you already like, or a full voice/style doc if you have one.]

Match that voice exactly in the rewrite. Where a query would only fit by sounding generic
or salesy, say so instead of forcing it in.

Read the export the way Claude should be prioritizing it

• High impressions, weak click-through rate, already on page one → the page is being seen and skipped. Usually a title/meta problem (Flavor B) or a weak opening line (Flavor A).
• Real impressions, position roughly 8-20 → “striking distance”: the page is already relevant enough to show up, just not developed enough to rank higher. A content-depth problem; add real coverage, don’t just sprinkle the keyword in.
• A query with real apparent demand that isn’t in the export at all, or shows near-zero impressions → the page genuinely doesn’t address it yet. That’s a gap, not a rewrite.
• High position and a healthy click-through rate already → leave it alone. Don’t “optimize” something that’s already working.

Real result

The payoff here isn’t a specific number, it’s what the reader can do without buying anything or learning a new dashboard: run the same optimization pass a paid SEO tool exists to sell, against data they already have free access to. Every input this pattern needs, which queries a page ranks for, how often it’s seen without being clicked, which ones sit just off page one, already lives in a Search Console account the moment a page starts earning impressions. This pattern turns that free export into a rewrite of the page’s own copy, in the page’s own voice, with no separate SEO subscription, no new login, and no agency in between.

Variations

• Which page first? Pull an unfiltered Performance export (skip the page filter in step 2) sorted by impressions to see which pages have the most untapped opportunity site-wide, then drill into those one at a time with the pattern above.
• Non-primary markets or devices. Search Console’s Performance report also filters by country and device. If a page serves more than one market, filter to the audience that actually matters before exporting, otherwise a large secondary-market query volume can crowd out the prioritization for your primary readers.
• A full voice doc instead of a few pasted examples. The voice-rules pattern covers turning a written vocabulary/style doc into a standing enforcement prompt (or a Claude Skill), the sturdier version of the tone extension above.
• Automating the export itself. This recipe works from a manual CSV download on purpose, it needs nothing beyond a Search Console account and Claude. A future recipe on this site will cover pulling Search Console data on a schedule into the traffic dashboard, the same repo-route pattern already used for PostHog traffic data.