--- title: "Your Rankings Held and Your Clicks Fell. Which One Actually Happened?" url: https://hostmy.blog/rankings-held-clicks-fell/ date: 2026-09-17 modified: 2026-09-03 lang: en author: "Aditya Sharma" description: "Four different problems produce the same Search Console chart. Here is the decision tree I use on client sites to tell them apart." categories: - "AI Search" image: https://hostmy.blog/wp-content/uploads/2026/09/hmb-card-1873-1024x538.jpg word_count: 1612 --- # Your Rankings Held and Your Clicks Fell. Which One Actually Happened? The message arrives about once a fortnight now, and it is always some version of the same three lines. Rankings look the same. Traffic is down. Have we been penalised. Four completely different problems produce that chart, and three of them have nothing to do with AI. Guessing wrong costs a month, because the fix for one is actively harmful for another. So here is the diagnostic my team runs, in the order we run it, with the checks that separate each case. ## Get one honest chart before you form a theory Almost every wrong diagnosis I have made started with a bad comparison. Open Search Console, Performance, Search results. Set the date range to compare two periods of equal length, matched to the same days of the week. Turn on all four metrics: clicks, impressions, CTR, average position. Then do the thing most people skip. Split brand from non-brand, using a query filter on your own brand name, and look at each separately. A brand query collapse and a non-brand collapse are unrelated problems, and averaged together they hide each other. Write down four numbers for each half. Now you have something to reason about. ## The four patterns, and what each one means | Impressions | Position | Clicks | Most likely cause | | ----------- | -------- | ------ | ----------------- | | Up or flat | Flat or better | Down | Answer served before your link. Zero click behaviour. | | Flat | Flat | Down, brand only | You lost a SERP feature, sitelinks or a knowledge panel | | Down | Down | Down | An ordinary ranking loss. Treat it as an ordinary ranking loss. | | Down | Flat | Down | Demand fell. Seasonality or a dead topic. | The first row is the one people arrive with and the one they most often misidentify, because it does not look like anything is wrong. It is the row where [ranking and being the answer come apart](/ranking-vs-being-the-answer/). The last row is the most embarrassing to miss. Traffic falls by a third, three weeks go into technical work, and then somebody checks Google Trends and finds the search term itself fell by roughly the same amount. Nothing about the site changed. The world did. Check demand before you check anything else. It takes two minutes and it disqualifies an entire class of theory. ## Separating position four from position four Average position is the most misleading number in that report, and it hides two things. First, it averages across every query the page ranks for, so a page can hold position 4 on average while losing its two best queries and gaining five poor ones. Look at the query-level view, sorted by lost clicks, rather than the page average. Second, position is measured against the ten blue links. It does not tell you what is above them. A page holding position 3 under an expanded answer block is in a very different situation from a page at position 3 with nothing above it, and Search Console reports both identically. That second point is what makes the first row of the table so common now. Your rank is a rank within a list that has been pushed down the page. ## The confounders that look like AI and are not Before concluding anything, rule these out. Each one produces a clicks-only drop. **Sitelinks disappeared.** Common after a navigation change or a slug change. Brand queries lose CTR immediately, non-brand is untouched. Search your brand and look. **A rich result stopped showing.** Recipe cards, FAQ, review stars, product boxes. Check the enhancement reports in Search Console for a date that matches your drop. **Your title got rewritten.** Google rewrites titles routinely, and a rewritten title can lose a chunk of CTR while position stays identical. Compare what you wrote to what actually shows. **A competitor arrived with a better snippet.** Nothing changed on your side. Someone else got more clickable. **Tracking broke.** Consent banner change, tag manager change, an analytics migration. Cross-check Search Console clicks against your analytics, because if only one of the two moved, the problem is measurement. That last one has fooled me twice, both times after a cookie banner update. Check it early. It is free. What Search Console can and cannot account for What Search Console can and cannot account for Search Console reports it Search Console has no field for it Impressions rising Clicks falling Average position held Answer shown above your result not reported Citation inside an assistant answer not in the API Both stories fit the same chart. Search Console reports the impression and has no field for the answer that consumed it. ## The check that confirms zero click behaviour If demand held, position held, impressions held or rose, and none of the confounders apply, then you are in the first row of the table. Two ways to confirm it rather than assume it. The first is to go and look. Take your five highest-impression queries, run each one in a private window, and note what sits above the first organic result. That is a slow, manual, entirely reliable check and no tool replaces it. The second is a sanity check on the ranking assumption itself. Ranking and being quoted are close to independent. In an analysis of 153,425 citations across six AI platforms in May 2026, only 23.05 percent of cited URLs rank in the organic top ten, and for ChatGPT that figure is 4.2 percent, which is why [the habits that stop working on ChatGPT](/chatgpt-seo-what-changes/) are a longer list than people expect. Ahrefs found the same drift on the Google side: across 863,000 keywords, AI Overview citations coming from top-ten pages fell from 76 percent to 38 percent. So a page can hold its ranking and lose the answer at the same time. Those are two different competitions with two different scoreboards, and the second one has no scoreboard you can see. The answer frequently goes somewhere you cannot install a plugin at all, which is [why Reddit threads and video transcripts get cited so often](/why-ai-cites-reddit-youtube/). The arithmetic of what this costs, worked through on a real client site, is in [what happens to a blog when the answer replaces the visit](/zero-click-search-cost/). ## What not to do when you find it The instinct is to publish more, and it is wrong here. New pages start at a median age of zero days, and the median cited page is 298 days old. Publishing into this problem is paying to start at the back of a queue you have not read the rules for. The second instinct is to add a file and call it fixed. llms.txt is the current favourite. SE Ranking looked at roughly 300,000 domains and found no correlation with citations, and removing the variable improved their model. Ahrefs checked 137,000 domains, found about 38,000 carrying the file, and 97 percent of those files were never requested once. Google says it does not use it. Add it if you want, it takes a minute, but do not mark the problem as solved. The third instinct, blocking the crawlers out of annoyance, is at least a real decision rather than a fix. It is worth making deliberately rather than by leaving a security plugin on its defaults, and the case either way is in [whether GPTBot has ever visited you](/gptbot-wordpress/). ## What actually moves the needle here Two things, both editing rather than plumbing. **Make the answer quotable.** The mean cited sentence in that dataset was 9.27 words, the median 10, and nothing above 18 words was cited even once. If your answer lives inside a 26 word sentence, there is nothing to lift out. **Move it up.** 41.9 percent of citations come from the first 30 percent of the page. On a fifteen hundred word post that is the first 450 words. Two structural notes that come free with the same edit. Sequential heading order, H2 to H3 to H4 without skipping, correlates with a 2.8x lift. Cut sections rather than adding them. Then maintain what you already have. Stale pages are roughly three times more likely to lose citations they held. Old and updated is the strongest state a page can be in. ## The measurement problem you should know about An honest caveat, because it changes what you can promise anyone. Gemini still exposes the exact quoted sentence through a text fragment on 84.1 percent of its citations, so on that one surface you can see precisely which line was used. Google AI Mode dropped that to 0 percent. Elsewhere you get no referrer and no confirmation. Which means the loop is broken. You cannot A/B test your way through this the way you could with title tags, and anyone selling you a citation guarantee is guessing with confidence, which is the first of [the AI SEO claims that do not survive contact with evidence](/what-ai-seo-means/). What you can verify is server-side. Whether the agents fetch you, how often, and which pages they take. That is a log question, and the method is in [checking which agents actually arrived](/read-server-log-ai-crawlers/). ## The WordPress side, once the diagnosis is done The editing above is yours. The machine-facing structure is repetitive enough that doing it by hand across a few hundred posts introduces errors. [RankReady](https://wordpress.org/plugins/rankready-ai-llm-seo/) is a WordPress AI SEO plugin that handles that half: summary blocks near the top of a post, FAQ and schema output, a Markdown version of every page for machine readers, freshness signals, and per-crawler control. It runs alongside Rank Math, Yoast, AIOSEO and SEOPress rather than replacing them, since your existing plugin still owns titles, metas and sitemaps. Setup is about five minutes. It makes pages fetchable and readable. Diagnosing which of the four problems you have is still a human job, and it is the part worth doing carefully. Before you change anything: have you split brand from non-brand yet, or are you looking at one averaged line and guessing?