Nineteen, twelve, eleven, ten: how to report a market with no leader

The four most cited hosts on the Gemini side of our 15 October 2026 run appeared in 19, 12, 11 and 10 of its 291 readings: webfx.com, zapier.com, semrush.com and margen.net. Four rows separated by nine readings in total. There is no leader in that table, and a report that prints it as a top four has already told its reader something false about the market by the choice of format.

This page is about the reporting rather than the result. A flat distribution needs a different template from a concentrated one, and almost every tool in this field ships only the concentrated template.

The four rows, and what the gaps between them are worth

Position	Host	Readings citing it, of 291	Gap to the row above
1	webfx.com	19
2	zapier.com	12	7
3	semrush.com	11	1
4	margen.net	10	1

One reading separates the second row from the third and the third from the fourth. A single question answered differently would reorder them, which means the order is not information. The levels are information, and the format has to show levels rather than places.

Five rules for reporting a table with no leader

• Print the count, always, next to the name. A name without its count invites the reader to assume a hierarchy that one reading would overturn.
• Print the denominator once per column. 19 of 291 is a reading. Nineteen is a number with no unit.
• Drop the word ranking. Call it the hosts cited most often and let the counts do the ordering, or group rows that sit within a reading or two of each other.
• Do not report a change in order between two readings as a change in the market. At these levels, order moves on its own. Report whether the levels moved.
• Say how many rows exist below the ones you printed. The whole run recorded 10,652 citations across 2,708 distinct domains, and four printed rows out of that is a selection that needs declaring.

Those five turn a table that misleads by format into one that can be read correctly by somebody who was not in the room when it was built.

What a flat table means for the work, briefly

It means presence rather than position is the thing to aim at and the thing to report. A business appearing in two of a hundred readings on a table like this is genuinely in the set, and the same two appearances on a table where one host holds most of the readings would mean something quite different. Same number, two different facts, and only the shape of the table tells them apart.

It also means your own first reading is the only sensible comparison for your second. There is no published average to be above or below when the top of the table is at 19 of 291, and borrowing a benchmark from a differently shaped surface produces a target that cannot be hit or missed meaningfully.

The same run, the other engine, for contrast

The other engine we read on 15 October 2026 produced a source table with its leading host in 176 of its 306 readings. That table has a shape a ranking describes correctly, because the distance between first and second is large enough to survive a question or two going the other way. So the right template depends on the surface, and a tool that applies one template to both has made a decision for you that the data should have made.

Where this reading stops

One engine's half of one run, 102 questions per market on 15 October 2026 across the United States, the United Kingdom and India, with one reading per question and 291 of those readings usable on that engine. Three of the five surfaces our product covers were asked nothing. The counts record a domain appearing in a source list, deduped to one per reading, and not a business named in the answer text. No per industry figure exists in the bank, and city level numbers come from a separate scan of ten named places.

We sell this measurement and the writing that follows it, which is why the unhelpful figure goes on the page with the rest: on that engine our own domain was cited 0 times in those 291 readings. The Google Search Console side of our own record, with both windows and the export date, is on our proof page.

What this does to a monthly report

The practical consequence lands in whatever document gets circulated every month, and it is usually the place a flat table does the most damage. A standing slide headed top five sources, with the names in order and no counts, will show movement every month on a table like this and none of that movement will mean anything. Somebody will then be asked to explain it.

Three changes fix that for good. Put the count beside every name so a reader can see that rows two to four are within a reading of each other. Add a line stating the denominator and the date the reading was taken. And replace the question of who is first with the question of whether we are in the set and how often, because that is the only field on a flat table that moves for a reason. A report built that way is shorter, harder to argue with, and does not generate a monthly mystery.

Questions about reporting a flat citation table

What were the four counts?

Nineteen, then twelve, then eleven, then ten, each out of the 291 readings that engine gave us on 15 October 2026.

Why does one reading between rows matter?

Because at that level a single question answered differently swaps two positions. The order is then an artefact of the sample rather than a property of the market, and a report that leads with the order has published the artefact.

Should I still name the top sources?

Name them, with their counts and the denominator, and without calling the list a ranking. The counts are the finding and the sequence is not.

How do I know whether my own table is flat or concentrated?

Count how many of an engine's sourced readings cited its own most cited host, and keep both numbers rather than the ratio. Two engines in our run came out at 19 of 291 and 176 of 306 on the same questions.

Which engines and date is this?

ChatGPT and Gemini, read on 15 October 2026. The method, the controls and the full source mix are published as 597 readings across ChatGPT and Gemini in three markets.

Choose the template after you have seen the shape

The sequence that goes wrong is choosing a leaderboard format first and then pouring any data into it. Look at the gaps between your rows before you decide how to print them. Where the gaps are one reading wide, group the rows and publish the counts; where the gap at the top is wide, a ranking is honest. The data decides the format, and the format is most of what a reader takes away.