Back to Blog

Search Demand vs. Visibility: Find the Trends You're Missing

Classify every topic as capturing, partially capturing, missing or losing search demand, using Google Trends and Search Console in Claude. A real audit, with the traps.

10 min readSEO & AEO

TL;DR

Paulo H. Alkmin shows a four-category audit (capturing, partially capturing, missing, losing) that compares Google Trends momentum, through his free trendzeist-mcp server, with Search Console impressions and positions inside Claude. On his own site, a comparison page lost 44% of its daily impressions while its position improved, because Google Trends interest in the topic fell from 100 to 2: a demand loss, not a ranking loss. He also found that unrelated keywords all stepped down on Google Trends in the same week of July 2026 while his Search Console data did not, and recommends checking for such step changes before trusting any growth percentage. His strongest gap was 'wordpress mcp', a topic with sustained demand and zero impressions on his site.

My comparison of n8n, OpenClaw and Claude agents lost 44% of its daily impressions between August and October. Over the same months, its average position improved by about three places.

Any dashboard would paint that page red. Google Trends tells a different story: interest in "openclaw" peaked at 100 in March and was at 2 in early November. The page didn't lose its ranking. The topic lost its audience, and the page is holding a bigger share of a much smaller crowd.

It's the most common misreading of a traffic report, and it sends people rewriting pages that are doing fine. The opposite case, a site slowly sliding while demand grows, rarely shows up on a dashboard at all, because a flat line feels safe.

The prompt below is built to tell those two cases apart.

Four categories, one decision each

The audit sorts every topic into one of four buckets, and the bucket decides the action:

CategoryWhat the data showsWhat to do
A. Capturing the trendYour impressions move with demand, or beat itProtect it. Refresh, don't rewrite.
B. Partially capturingGoogle shows you, but on page three, or with a CTR far below the positionImprove the page that already ranks
C. Missing the trendDemand exists, your impressions are zeroWrite one page, if you can credibly answer the query
D. Losing despite demandDemand flat or rising, your impressions and position both fallingFind out why before writing anything

What separates them is a comparison of rates. Google Trends gives you the percentage change in interest. Search Console gives you the percentage change in your impressions. When the two move together, you're category A even if both lines go down. When they split, you have a problem or an opportunity.

Average position is the tiebreaker. Falling impressions with a steady position is demand. Falling impressions with a falling position is you.

The prompt

Identify emerging search topics where external demand is growing faster
than my website's visibility.

Use Trendzeist to find breakout and rising topics in my site's subject areas.

For each topic:
- determine its trend direction and momentum,
- search Google Search Console for related queries,
- determine whether impressions and clicks are keeping pace with the trend.

Classify each topic as:
A. We are capturing the trend
B. We are partially capturing the trend
C. We are missing the trend
D. We are losing visibility despite increasing demand

Focus especially on category C and D.
For each one, recommend the specific SEO action I should take.

Compare the same date windows in both sources. Use average position to
separate a drop in demand from a drop in ranking. Before computing any
growth rate, check whether all keywords stepped up or down in the same
week, and say so if they did.

The last paragraph is my addition, and it exists because of what happened the first time I ran this.

The week everything fell

Here is a slice of the 12-month curves Trendzeist returned for five keywords that have nothing to do with each other:

KeywordWeek of 5 JulyWeek of 12 July
claude code5939
claude code skills7137
wordpress mcp7242
woocommerce subscriptions4222
rag chatbot6238

Five unrelated topics lost a third to a half of their interest in the same seven days, and none of them recovered to its old level. Real demand doesn't move like that. When unrelated lines fall together on the same week, suspect the instrument before you suspect the market.

I can't prove what changed in Google's data that week. I can say that Search Console didn't see it: my daily impressions for Claude-related queries had the same average in the week before 12 July as in the week after.

This matters because Trendzeist's growth_3m compares the mean of the last three months with the three before, and from early November that window straddles the step. Part of the 49% drop for "claude code" I quoted in the previous post sits on that week. The direction holds, since the curve kept sliding after the step. The size of the drop is softer than the number says.

So the rule: before you trust a growth percentage, look at the curve. Trendzeist returns the weekly points along with the summary for exactly this reason. Since version 0.6 it also checks each series for a sustained step, reports it as level_shift, and adds a shared warning when several keywords in one call step in the same week. Run on three of these keywords together, Trendzeist now flags the week on its own: one level_shift warning dated 12 July, shared by all three series, marked as a likely artifact. Read the curve anyway. A detector only flags the steps it was built to see, and one call normalizes all its keywords together, so a big keyword like "claude code" can flatten small ones to a row of zeros. Compare keywords of similar size in the same call.

The run on phalkmin.me

Claude ran discover_topics and interest_over_time for the topics my site covers, then pulled 90 days of Search Console data by query and by page. This run is from 3 November 2026, and I've rounded my own Search Console figures into bands. Nine topics came back:

TopicTrendMy visibilityCategory
claude codeDown 49% over 3 months (partly the July step)Claude-query impressions roughly doubled from early July to late October, lower half of page oneA
openclaw100 in March, 2 nowComparison page: daily impressions down 44% from August to October, position up about three placesA
n8n consultant"n8n" down 49%Several hundred impressions on page four, no clicksB
rag consulting"rag chatbot" down 61%A few hundred impressions across three variants, page threeB
claude code skills100 in May, 20 nowPost: a trickle of impressions, more in Portuguese than in EnglishB, and the wave already passed
woocommerceRising now, +26% over 12 monthsTroubleshooting posts on pages three to sixB
llms.txtPeaked in June, down 63%, 12 news articles in 30 days0 impressionsC
generative engine optimizationDown 34%, 12 news articles in the last 7 days0 impressionsC
wordpress mcpPeaked in June, down 54%, back to its 2025 level0 impressionsC

No D. The prompt pushed hard for one, and the Search Console server's content_decay tool, which looks for three consecutive months of decline, came back empty. Most of what looks like a D is a category A page attached to a dying topic, like the OpenClaw comparison. Check position before you call it.

What to do with the C's

A C is only an opportunity when you can answer the query better than the pages already ranking. "Is there demand?" is the easy half. The hard half is whether your experience and your business actually belong in that query.

On my list, the three C's get three different answers.

wordpress mcp: CREATE

The strongest of the three. Demand fell from its June peak back to where it was a year ago, which means it never went away, and it sits exactly on two things I sell: WordPress AI integration and custom MCP servers. Somebody searching "wordpress mcp" is often a developer or an agency trying to connect an AI assistant to a client site, which is a buyer for one of those two services. Action: CREATE one practical page, built from my own setup, and link it from both service pages on day one.

llms.txt: CREATE, shaped for answers

A format question with a specific audience. Search interest dropped, but Search Engine Journal, Search Engine Land and Search Engine Roundtable published 12 articles about it in 30 days, and the questions people type ("how to create llms txt file", "llms.txt vs robots.txt", "why llms.txt not working") have short, citable answers. This site generates its own llms.txt from content at build time, and ParseLess serves Markdown to AI crawlers. Action: CREATE one answer page shaped for AI citations, opening with a 40 to 60 word answer under each question heading, which is the worked example in the AEO research post.

generative engine optimization: IGNORE as its own page

It has news momentum (12 articles in a week), and Wikipedia already has an article on it. Competing on "what is GEO" means competing with Wikipedia and every SEO publication at once. Action: IGNORE as a standalone page. Answer the comparison question ("GEO vs AEO vs SEO") inside an existing SEO post, where it supports something I actually do.

What to do with a real D

I didn't have one this time, but the steps don't change:

  1. Confirm it's real. Demand flat or rising across the same window, with no step change, and your position falling along with your impressions.
  2. Check whether another page took it. The Search Console MCP server has a cannibalization_check tool for this. Two of your own pages splitting one query looks exactly like a ranking loss.
  3. Look at the results page for the query yourself. A new AI Overview, a video carousel or a forum block can push your result down without anything on your page changing.
  4. Only then refresh the page, with the specific thing the current top results cover that yours doesn't.

A new post comes last. Most D's get fixed on the page that's already ranking.

When demand moves instead of disappearing

One more check before calling any topic dead. Trendzeist's compare_channels splits interest by Google surface, and for "claude code" it returned this:

SurfaceChange, last 3 months
Web search-49%
YouTube-8% (stable)
News+213%

Interest in Claude Code moved toward news coverage, while video held steady. I can only guess at the reasons, and the action doesn't depend on them. For a tutorial site that's a signal to keep the evergreen guides fresh, put new effort into timely pieces, and stop treating web-search interest as the whole demand.

Run it every month

This audit is the one I'd run monthly on any site where organic search brings leads: Trendzeist plus the Search Console server, the prompt above, and an hour reading the table before anyone writes a line. The setup takes three commands.

If you'd rather hand it over, the same audit, with Analytics added and every row checked by hand, is my Search Opportunity Audit. To talk it through first, book a free 30-minute Intro Call.

Which of your falling pages is actually doing fine in a topic that's shrinking, and which flat one is losing ground?

Tags

  • Trendzeist
  • Google Search Console
  • MCP
  • SEO
  • Content Strategy

Frequently Asked Questions

Read More Posts

Explore other articles and insights

Back to Blog