Seventeen of the Ranked Lists on Page One Put Their Own Publisher First

Search for the best tool in this category and Google returns a page of ranked lists. We took four ways of asking, read the organic top 20 of each, and opened every ranked list on it. Seventeen of them are published on a domain that sells one of the tools in the list, and in all seventeen that tool is at position one.

Fourteen distinct companies. Two of them appear under more than one wording of the question, so the same self-ranked article set covers several ways a buyer might phrase it.

None of this is hidden or against any rule. It is ordinary content marketing, and most of these pages are useful. The finding is about what a results page is made of when you ask it who is best, and about what happens next: AI answers for the same question cite some of these pages as sources.

Disclosure: EchoWi sells in this category, several companies named below are competitors, and we publish a comparison of the market too. Ours is sorted by entry price rather than ranked, we appear in it in the second row on that sort with the lowest engine count in the table, and it carries a section on when we are the wrong choice. Everything needed to repeat this against us is here: the four queries, the market is the United States in English, the date is 11 August 2026, and every row sits in our measurement register with the publisher, its own tool and its rank.

The short version

1. Seventeen ranked lists in four top-20s put their publisher’s own tool first. Fourteen distinct companies.
2. We opened every one. Google’s snippet was right seventeen times out of seventeen; our own detector was wrong twice, which is the more useful half of that sentence.
3. The category’s own AI answer cites some of these pages. Asked which tool is best, the stable sources include a domain whose ranked list places itself first.
4. This is not a scandal and it is not nothing. A buyer reading page one is mostly reading vendors ranking themselves, and nothing on the page says so.
5. The countermeasure is cheap. Check who owns the domain before reading the ranking. It takes one look at the footer.

How it was measured

Four ways of asking the same commercial question: best AEO tool, best geo tools, best ai visibility tools and best llm monitoring tools. Four separate wordings on purpose, because we have measured before that rewording a question changes which sources come back, and a finding that only holds for one phrasing is a finding about the phrasing.

For each, the organic top 20 in the United States in English on 11 August 2026. Every result that is a ranked list of tools was opened, and two things recorded: whether the publishing domain sells a tool in that category, and whether that tool sits at position one.

The fourth query is a deliberate control from an adjacent market. LLM monitoring in the developer sense is observability tooling, a different category with different companies, and it behaves the same way: four of its ranked lists are published by a platform sitting at its own position one. Whatever this is, it is not specific to AI visibility.

What we do not print, and why

Every one of these pages describes its own product in superlatives. We are not going to quote any of them.

The reason is practical rather than delicate. Reproducing a competitor’s marketing sentence on our domain restates it, and a passage lifted out of context by a retrieval system arrives with our name attached to their claim. It cost us a correction in August, on a page where the quote was accurate, attributed and in a paragraph criticising it.

So the recorded fact is structural and checkable without our help: this domain sells a tool, publishes a ranking, and occupies position one in it on this date. Anyone can open the page and confirm or refute it. Nothing about whether the ranking is deserved is claimed here, because we did not test the products.

Opening the pages changed two rows

Every row was verified by loading the article and reading its first list position rather than trusting Google’s snippet. That step exists because a snippet is Google’s extraction, and we wanted our own.

It nearly went wrong in the opposite direction to the one we expected. A small script looking for the first 1. on the page returned the wrong item twice, because both pages carry a numbered list of evaluation criteria above the list of tools. Read literally, our detector said those two publishers did not rank themselves first. They both do, in the tool list, exactly as the snippet said.

So on this measurement Google’s extraction was right seventeen times and ours was right fifteen. The lesson is the one this site keeps relearning: a detector defines its own universe, and it will confidently return the wrong answer rather than no answer. The correction cost two page loads.

Why this matters more than it used to

A ranked list written by a participant is an old genre and readers have old defences against it. The reason to measure it now is that a second reader has appeared that has none.

Asked the buying question for this category on Google’s AI surfaces, with the cache bypassed and repeated runs, the sources cited in every single run include a domain whose own ranked list puts its own product at position one. The engine does not present it as a vendor’s list. It presents it as a source.

That is the mechanism worth understanding, and it is the same one we found measuring who holds the durable citation slots across eleven categories: the slot goes to whoever publishes the comparable thing, and a vendor publishing a ranking is publishing exactly that. Being ranked by an intermediary is hard. Becoming the intermediary is a page.

It also lines up with the other half we measured this week. Programmatic “A vs B” comparison pages mostly rank for nothing, while a ranked list of the whole category does rank, and does get cited. The category list is doing work the pairwise pages are not.

What a buyer should actually do with this

Check the footer before you read the ranking. Every one of these seventeen pages is transparent about who owns the domain if you look. None of them announces it next to position one. That asymmetry is the whole issue and it is resolved in about five seconds.

Read the criteria, not the order. The useful part of a vendor-written list is usually the comparison table: prices, engine counts, limits. Those are checkable against the vendor sites and they are often accurate. The ordering is the part that carries the interest.

Count how many lists agree. Across these four queries, the tools that appear repeatedly in other companies’ lists are a different set from the tools that appear at position one in their own. The first set is closer to what the market thinks.

And if you are buying because an AI recommended it, ask which sources the answer used. If the recommendation traces back to a ranking published by the recommended vendor, you have learned something specific about that recommendation.

What this means for our own list

We publish a comparison of this market, so the finding applies to us and it would be cheap to exempt ourselves.

Ours is sorted by entry price. That is not a modest ranking, it is not a ranking at all: the order is arithmetic and anybody can verify the sort. On that sort we sit in the second row, which is a fact about our price and not about our quality, and we carry the lowest engine count in the table. The disclosure sits above the table rather than in a footer, and there is a section on when we are the wrong choice.

We are not claiming that makes it neutral. A company comparing a market it sells into has an interest whatever the sort order. It does mean the ordering carries no claim, which is the specific thing the seventeen pages above do carry.

There is a second problem with our list that the sort order does not touch, and we found it by measuring rather than by auditing ourselves. Putting the same buying question to an assistant instead of to Google returns a set of cited domains, and 17 of the companies cited that way sell exactly this software and are not in our catalogue. A list maintained by hand lags a category that is still forming, and ours lags it by about a quarter.

The same design in Spain, where it does not transfer

Everything above is the United States in English, which the limits section called this study’s largest weakness. So we ran two Spanish queries in Spain the same way: mejores herramientas geo and herramientas aeo, organic top 20, every ranked list opened.

The difference is not that Spain has fewer of them. It is that two of the four behave differently. One vendor’s list names other tools and places its own product fourth in the sentence; another publishes a comparison of the market in which its own product does not appear in the ordering at all. In the United States sample there is no such page: every vendor-published ranking found there is a vendor-published ranking with that vendor first.

What fills page one instead is agencies explaining the concept. For herramientas aeo, most of the twenty results are consultancies and marketing agencies answering what AEO is, not rankings of tools. That is what a category looks like earlier: the demand is still definitional, so the content that ranks is still definitional, and the tool-ranking land grab has not happened yet.

And the Spanish keyword has a problem the English one does not. mejores herramientas geo returns the Spanish open-data portal writing about geospatial visualisation tools, and a social post introducing a series about the tools geographers use. In Spanish, geo is doing double duty, so a slice of that results page is not our category at all. This is the same trap our own research rules already carry for brand names, showing up in a category term: the volume is real and a chunk of it belongs to somebody else.

So the honest scope of the headline is a market and a moment, not a category. Where the buying question is still being asked as a definition, page one is written by people explaining. Where it is being asked as a purchase, page one is written by the sellers.

And Germany, which is a third shape

Two German queries, same design. beste geo tools and ki sichtbarkeit tools, organic top 20, every ranked list opened.

Germany has the self-ranking layer that Spain mostly does not, and it has something neither of the others has: a dense layer of local agencies publishing rankings of other companies’ tools. Marketing agencies, a media house, a social-media academy and several consultancies all rank the same international vendors, and none of them sells one. In the American sample there is no such page; every ranked list found there was published by a participant.

So the three markets are not one pattern at different sizes. The United States is vendors ranking themselves, Spain is agencies explaining what the category is, and Germany is agencies ranking somebody else’s tools while international vendors rank themselves on the same results page.

The German number is a floor and the reason is worth stating. Three more candidates were dropped rather than counted. One page returned no body to our request at all. Two carry no numbered list, so their ordering could not be read the way every other row was read. Google’s snippet suggests at least one of the three does place itself first, and a snippet is not the standard the other twenty-three rows were held to, so it is excluded rather than counted.

France completes the picture, and it is the opposite of the United States

One French query, meilleurs outils geo, same design. In the whole organic top 20 there is one vendor-published ranking that places its own tool first. Everything else is French agencies, consultancies and a comparison site ranking other companies’ tools.

Read together, the four markets say something a single one could not. The ranking layer for this category is owned by the vendors in the United States and by the agencies in Europe. In America the page that tells a buyer who is best is written by sellers; in France it is written by the people who would implement the tool for them, and in Germany by both at once.

That is a commercial fact rather than a moral one, and it points in opposite directions depending on where you sell. In the United States, being ranked means being ranked by a competitor. In France it means being ranked by a potential partner.

One French page in that top 20 is worth a footnote. It is machine-translated, and it rendered the category term as géocroiseurs, which is French for near-Earth objects. It therefore recommends a platform for tracking asteroids. We confirmed it on the page. It is a curiosity rather than a data point, and it is the third form of the same problem this study has now hit three times: an acronym with other meanings, a short root that collides in another language, and a translation that resolves the wrong sense.

Do the engines cite that European agency layer?

The sweep above says who owns the results page. The next question is whether an assistant reaching past it cites the same people, and Spain and Germany answer differently.

In Spain four of the six domains present in all three runs are Spanish marketing agencies or an SEO consultant: cyberclick.es, eleven.agency, ethinking.es and sergiovazquez.es. The structure of the results page carries into the answer.

France repeats it exactly. Four of its six stable domains are French agencies: Eskimoz, WAM, Natural-net and Nocode Factory, each of which calls itself an agency on its own front page and none of which sells a tool in this category. The other two are YouTube and LinkedIn.

Germany does not. The domains present in all three runs are YouTube, HubSpot’s German site, SE Ranking and one vendor. Every German agency we saw dominating that results page appears in one run of three and never in all of them.

France was run to test a prediction, and it is worth saying that it passed. Having found the French results page the most agency-dominated of the four, we wrote down beforehand that its answer layer should look like Spain’s rather than Germany’s. It does. We note it because four tidy hypotheses died in this research in two days, and a study that only reports the predictions that came true is not reporting predictions.

And then a fifth died, which is the German row above. Reading three markets and finding Germany the exception invites an explanation about the German market, and we went looking for one. The first thing to check was whether the result belonged to the market or to the phrasing, so we asked a second German question: not which GEO tool is best, but how to measure whether your brand is visible in ChatGPT and Google AI. Two German agencies then held across every answered run, 40komma6.de and hechtinsgefecht.de, each describing itself as an agency on its own front page.

So there is no German exception to explain. The zero belonged to a wording, not to a country, and had we published the three-market table without asking a second German question we would have shipped a national explanation for a phrasing artefact. That second run answered twice rather than three times, which makes it unusable for comparing counts with the rows above and entirely sufficient for what it is used for here: killing a claim that said no local agency holds in Germany at all.

The first version of this section could not say that, and the fix is worth describing. Spain was originally measured across two surfaces and Germany on one, because the German call with two surfaces stopped after a single run and one run makes every cited domain stable by construction. Two designs cannot be compared, so that pair was published as a direction and explicitly not as a comparison. Spain was then re-run on the German row’s design, one surface until three answered, and the table above is that pair.

Matching the design did not soften the finding, it sharpened it: four of six against none of four, where the mismatched pair had read two of eight. The original Spanish row stays in the measurement register alongside the matched one, and so does a note naming the German attempt that was discarded, because an absence that looks tidy is the kind nobody questions.

Limits

Nine queries, four markets. Four in the United States in English, two in Spain, two in Germany and one in France, all on one day. One or two queries per European market are enough to show the shapes differ and not enough to characterise any of them, and France rests on a single query.

Top 20 only. Results below rank 20 were not read. A list on page three that ranks its publisher first is not counted here and there are certainly some.

“Sells a tool in the category” is a judgement. It was made by opening each domain’s own homepage rather than by recognising the name, which is the method that has twice returned a different answer from our first guess. Two domains in these results are agencies rather than tool vendors and are not counted.

No quality claim. Nothing here says a self-ranked list is wrong about its ordering. We did not test the products, and a vendor can be both interested and correct.

Nothing is causal. No page was changed and nothing was measured before and after.

The comparison group this study never had

Everything above counts one kind of page: the ranked list published by a company that sells one of the tools. It never counted the other kind, and the register never stored it, so “seventeen do this” had nothing to be seventeen against.

On 19 August we went back and counted both. Two things change at once and both are said out loud: the scope grows to include non-sellers, and the date moves eight days, so page one has moved. This is a fresh reading, not an upgrade of the seventeen, which keeps its date and its method. Three of the four original queries were re-read; the fourth was the adjacent-market control and nothing here describes it.

31 ranked lists, 26 readable, 2 behind a bot challenge and 3 whose ordering this sweep cannot extract.

16 are published by a company that sells a tool in the category. All 16 are in their own list, and 14 put themselves at position one.

10 are published by somebody with nothing to sell in the category. Not one of the 10 names its own publisher. A headless web platform, an automation platform, a paid SEO community, an individual consultant and six marketing agencies each publish a ranking of AI visibility tools, and none of them is in it.

That is the comparison the original headline needed and did not have. It is also the second category where it holds: the agency arm found the same split across four markets, seven non-seller publishers and zero self-mentions.

The two who rank a competitor above themselves

The 11 August count was seventeen out of seventeen at position one. Opening these pages eight days later, 2 of the 16 sellers are in their own list and not at the top of it, and both put the same competitor first before naming themselves second.

That is worth saying plainly because it cuts against the easy reading. A vendor list is not automatically a vendor putting itself first; it is a vendor deciding where to put itself, and two of them decided on second. The agency arm turned up exactly one of these, a publisher at position three of seven, so this is now three across two categories.

And three lists could not be read in order at all

Three seller-published pages present their tools without an ordered structure this sweep can extract: comparison tables, tabbed panels, cards. Their publishers are in them; where is unknown.

They are counted as unreadable and not as zero, because a page that does not rank in a way we can read is a fact about the page and about our reader, not about the publisher’s modesty. The same distinction this study draws between a blocked page and a missing one.

Common Questions About Best-Tools Lists

Is it wrong for a vendor to publish a ranking that includes itself?

No, and this study does not argue that. It is standard practice, most of these pages carry real comparison data, and a vendor can be interested and accurate at the same time. What is worth knowing is the base rate: on these four results pages, seventeen ranked lists are written by a participant sitting at position one, and the page gives a reader no signal of that next to the ranking.

How do I tell whether a list is written by a vendor?

Look at the domain and its footer, then check whether the tool at position one shares the domain name. That catches almost all of them, and it took us five seconds a page. The harder case is a vendor whose product name differs from its domain, which is why we opened every homepage rather than reading names.

Do AI assistants cite these lists?

Some of them, yes. Asking the category’s buying question on Google’s AI surfaces with the cache bypassed and repeated runs, the sources that appear in every run include a domain whose own ranked list places its own product first. The answer presents it as a source rather than as a vendor’s list.

Does this happen outside AI visibility tools?

It did in the one adjacent market we checked. LLM observability, a developer tooling category with different companies, returned four ranked lists published by a platform at its own first position. One control is not a survey, but it argues against this being specific to this category.

Does EchoWi publish one of these?

We publish a comparison of the market, sorted by entry price rather than ranked, with the disclosure above the table and a section on when we are the wrong choice. On the price sort we appear in the second row, and we carry the lowest engine count in that table. A price sort still comes from a company with an interest; what it does not do is put a claim in the ordering.

What should I read instead?

Lists published by people who sell nothing in the category, and the comparison tables inside vendor lists rather than their orderings. Prices and engine counts can be checked against the vendor’s own pages in an afternoon, which is what our catalogue does and what anyone can do without us.

Where this leaves you

The result is not that these pages are dishonest. It is that page one for this question is largely a market talking about itself, and that the newest reader of that page, a retrieval system answering a buyer, has no way to tell a vendor’s ranking from anybody else’s.

There is a second half to this, measured later: when an assistant reads one of these vendor written lists, it does not necessarily pass the publisher’s own ranking along. We counted how often an answer that cites a vendor’s page also names that vendor, and being named and being cited turn out to be two different outcomes.

That is checkable in five seconds per page and almost nobody does it, including us until we counted. Seventeen out of the ranked lists on four results pages is a big enough number that the check is worth making a habit.