Ranking is not being named

Most content tools have an answer to this question, and the answer is that you do not need to worry about it. Rank well, they say, and assistants will name you, because assistants read search results. Outrank makes that case in public: 750,000 articles published, and 10,000 “ChatGPT mentions secured” advertised on the same page.

Ask how the mentions were counted and the answer is that they were not. No prompt is tracked anywhere in that product. There is no before, no after, and nothing tying a published article to a brand getting named. It is a claim about an outcome, sold by a tool that cannot see the outcome.

The annoying part is that it might be right. Ranking probably does help. But “probably helps” is a hypothesis, and you already know what this job does to hypotheses nobody measures. We spent fifteen years learning to distrust exactly this shape of argument about links.

A results page is a list. Ten links, plus whatever Google has packed in around them, and the person choosing gets to see everybody. Position four is worse than position one, but it is still a place to be seen.

An assistant answers with a paragraph. It names two or three products, occasionally five. There is no fourth place and no second page. You are in the sentence or you are not, and the person asking usually never finds out there was a list.

The question changes shape too. Nobody types “budget tracking app” into an assistant. They type “what should I use to track a freelance budget if I hate spreadsheets”, or “alternatives to X for a two person team”. Those are prompts, and not one of them has a volume number attached. So the first real task is deciding which twenty or thirty prompts stand in for the way your buyers actually ask. Get that wrong and everything downstream is noise dressed up as data. It is the least glamorous piece of this work and the one most likely to sink it.

Where the same page wins one and loses the other: none of this means the two are unrelated. Pages that get cited are usually pages that rank. The cases worth studying are the ones where a page ranks perfectly well and still never gets named, because the causes there are specific and mostly boring.

The answer is in paragraph four

Ranking tolerates a slow start. A page can sit at position five with four hundred words of throat-clearing in front of the useful sentence, because the link still gets clicked and the reader still scrolls. Assistants quote pages that answer directly and early. If your introduction sets context for three paragraphs before it says the thing, the sentence that gets quoted is on somebody else’s page.

Everything cited has a table and you have prose

Open the pages cited for a prompt where your competitors get named. They are comparison pages, alternatives pages, and “best X for Y” lists, structured so an answer can be lifted out of them whole. If every cited page for a prompt carries a comparison table and yours carries three paragraphs of nuance, it is not the nuance losing. It is the shape.

Your name is nowhere near the answer

This one shows up on genuinely good pages. You answer the question, you answer it early, and you never put your own brand beside the answer. The assistant takes the explanation and credits the topic. You wrote the source and somebody else got named in the sentence.

Nothing could read the page in the first place

Before rewriting a word, separate two failures that look identical from the outside: the content lost, or nothing could read the page at all. Whether the content arrives server-rendered, how the page behaves on redirects and errors, whether the document has real structure, whether the controls do anything to a machine that cannot click. A page failing there does not have a copy problem yet, and a rewrite will not touch it.

Prices checked at the start of September 2026. Ahrefs tracks 5 prompts on the $129 Lite plan, 10 on Standard at $249, and 20 on Advanced at $449. Brand Radar on its own starts at $199. Prompt packages sit on top of all that at $50, $100 and $250 a month, with overage between $0.010 and $0.020 a check. Semrush rebuilt its plans around the same thing: the $139 plan tracks no prompts at all, 50 a day starts at $199, and 200 a day is $549.

Now the other side. Running one prompt against one assistant and recording who gets named costs $0.0002 to $0.0006 in data fees. The money is in the model call underneath, and inside that it is the web search that dominates: OpenAI and Anthropic both charge about $10 per 1,000 searches, which is roughly $0.010 of a $0.012 check. Perplexity lands near half that. Blended across the three, a check costs a little over a cent.

So 25 prompts, across three assistants, checked every week, is 323 checks a month and under four dollars of data. Five tracked prompts at $129 a month is not a technical ceiling. It is a demo with a subscription attached, and it is why most people reading this have never seen more than a sample of their own answer coverage.

None of that makes Ahrefs a bad tool, and Beseen vs Ahrefs is specific about the parts where it wins outright. It also puts the two sets of prompt-tracking prices next to each other, plan by plan.

You do not need a tool for the first pass. You need an afternoon and a spreadsheet.

• Write 20 to 30 prompts the way a buyer asks, not the way a keyword tool phrases things. Cover your category, your two or three real use cases, and an "alternatives to" prompt for every competitor you care about.
• Run each one in ChatGPT, Claude and Perplexity. Record which brands get named and which URLs get cited. That is your starting point.
• For every prompt where a competitor is named and you are not, open the page that got cited and read its first hundred words. Then read yours.
• Sort your own pages by Google position before you change anything. Positions 1 to 3: leave them alone, a rewrite risks something that already works. 4 to 15: this is where a title and an opening can genuinely move. 16 to 50: you are missing a section, not a keyword. Past 50: you need a different page, not an edit.
• Re-run the same prompts, worded identically, a week later and a month later. One answer is an anecdote. The same prompts run repeatedly are evidence.

One thing to know before you start: citations do not move in days. If you rewrite on Tuesday and re-check on Wednesday, you have measured your own impatience. Give it weeks, keep the prompt wording frozen, and change one thing at a time on the pages you touch.

The tools split cleanly down the middle. Ahrefs, Semrush and Profound measure. Outrank, SEObot and Byword write. Profound will tell you precisely which prompts name your competitor and hand you no fix. Outrank will publish thirty articles a month and cannot tell you whether one of them changed a single answer.

Both halves are real work. Neither one closes. The full cycle is: find the prompts naming someone else, work out why the cited page wins, write the fix, apply it, then run the same prompts again and see what moved. That last step is what turns a claim into a result, and it is the step almost nobody sells you.

That cycle is the whole of what we built, and how it works covers what gets pulled, how often, and what a rewrite looks like when it lands. If you want the cheap version of step one instead, the free scan runs five prompts against one assistant and checks whether an agent can read your site at all. No signup.