75 citations on a query my site does not appear on at all
I trusted a number for about three weeks and it turned out to be two different numbers wearing one label.
Bing Webmaster Tools has a grounding query table. It lists queries and a citation count. Mine showed 75 citations and 30 percent share for the query AI multilingual support agents US ecommerce. I had been reporting that as one of my pages winning that query.
Search Console has a separate AI features report, which gives citation counts per URL. Different provider, different definition of a citation. I had been reading the two together as though the 75 belonged to a page in the second report. Nothing ever joined them.
So I ran the query in Bing by hand. The top five back were alhena.ai, sobot.io, robylon.ai, agent.nexus and fin.ai. My domain is not on that page anywhere, not in the answer block and not in the organic list.
The two pages I had assumed were carrying the 75 do not appear in the per URL report either. One of them uses the word multilingual eleven times and the word ecommerce zero times. The other one is the reverse, fifteen mentions of ecommerce and none of multilingual. Neither of them says US anywhere. So the query I thought I owned is being answered by two halves that never reference each other.
Running the query took two minutes and I had not done it once in three weeks. The rule I am keeping is that a number from one provider cannot be explained using a number from another until I have run the thing by hand at least once.
What is a metric you reported for weeks before you checked what was underneath it?
Replies
Mine was an average position, and for two weeks it was the only good news in the whole Search Console export.
One template page sat at position 5, impressions climbing, zero clicks. I wrote it down as a title and description problem: we rank, nobody clicks, so the snippet must be weak. That reading felt obvious, and the pages export supports it perfectly, because the pages export and the queries export do not join. You get a position per URL and a position per query, and nothing tells you which query belongs to which URL.
Filtering the report to that one page took a minute. The query it ranked for was the name of the company on its own sample invoice. The template is filled with an invented seller so the preview looks real, almost nothing else on the web carries that string, so we came top five for it. A second page did the same with its invented buyer. Every non-brand top ten position the site had was one of these. There was no commercial query under any of them to write a better title for.
The other one ran the opposite way. Analytics said zero downloads from our free generator, and that zero was true, but not for the reason the dashboard implied. The server log had nine downloads, all of them ours, and two of those were from an iPhone that analytics never recorded at all, most likely because the event fired after the download had already taken the page away. So the zero was right about visitors and wrong about the tool, and I only knew which half to believe because the log disagreed with it.
The rule I kept is close to yours: an aggregate is not evidence until I have opened the row underneath it, and a count from the client is not evidence until something on the server agrees.
Your two halves that never reference each other is the part I will steal, though. That is a content fix, not a measurement one, and it is the harder of the two.
@alex_goldwyn The invented seller name ranking top five is a great catch. No snippet rewrite would ever have touched that, since nobody was searching for invoices in the first place.
Neither of my two pages has changed yet. What I keep going back and forth on is whether one page should carry both words, or the two should just link to each other and let the query sort it out. Did you do anything with those template pages, or just stop counting them?
@abdullah_javaid3Â Just stopped counting them. The market research page is unchanged, it still ranks for its own sample seller, and there is no commercial query under it to write a title for.
On one page or two, our data points one way. We have a staffing agency template plus separate pages for nurse staffing and recruitment agencies. Google matches the whole cluster, nurse and recruitment queries included, against the one staffing page, and the two dedicated pages get one or two impressions each. So the work went into that one page: six more answers, each written against a query the export actually shows.
And we have your exact case running right now. A how-to on the same subject went live at the end of August, started pairing with the main page on the same queries, and the main page dropped. The drop does not say why, because a site-wide dip and a title rewrite landed in the same weeks. So both are frozen until 28 September, and if the main page has not recovered by then, the how-to gets redirected into it. I will come back and tell you how it went.
@alex_goldwyn The nurse and recruitment pages getting one or two impressions each is the answer I was missing. If Google already folds the whole cluster into one page, splitting multilingual and ecommerce across two was always going to leave both half matched.
Freezing both until the 28th is the part I'd have skipped. How are you telling the site wide dip apart from the title change when the date comes, just the main page against the rest of the site?
@abdullah_javaid3Â Honestly, I am not telling them apart, and I stopped trying. Three things landed on that page within two weeks: rewritten titles across the whole family, the how-to page appearing beside it, and an edit to the page itself. The same week the main page fell, guides and comparison pages fell by similar proportions. At fifty to a hundred impressions a day, no attribution survives that.
So the decision was written as a threshold on the outcome instead of a verdict on the cause. For the week of 21 to 27 September: if the main page is back to at least 100 impressions at an average position of 50 or better, both pages stay. If not, and the how-to is still splitting the same queries, it gets merged into the main page with a redirect. Either answer is fine whichever change caused the dip, which is the point.
The rest of the site is the only control I have, and it is a weak one. The lasting lesson is the process one: one mass change per page family at a time, then four weeks of nothing, because these pages get recrawled only every few weeks. I will come back here with the number after the 28th.