Best AEO tools: three of six pricing pages carried no price a crawler could read

What do the best AEO tools actually do?

Two products are sold under the one label, and they answer different questions. The first is an answer tracker. It holds a list of prompts, sends them to a set of models on a schedule, and counts how often the answers name your brand, which competitors they name instead, and which pages they cite. Profound, Peec AI, Otterly.AI, PromptAlpha, AthenaHQ, the Ahrefs Brand Radar product and the Semrush AI Visibility Toolkit are all this machine, whatever each calls its metric. The second is a read-side scanner. It fetches your pages the way a crawler does and reports whether the words, the links and the markup survive that fetch.

The difference is not a feature comparison, it is a difference in what can be known. An answer tracker observes an outcome and cannot tell you why it moved, because the model is not obliged to explain itself and its answer is not stable between runs. A read-side scanner observes a cause and cannot tell you whether fixing it changed any answer. Lantad's own measurement of prompt stability found that repeated runs of the same prompt kept moving for several rounds before settling, which is the honest reason a single week of tracking data proves less than it looks like it does.

Buying the wrong machine is the expensive mistake in this category, and it is easy to make, because both are marketed with the vocabulary of generative engine optimization. If nobody can fetch your pages, an answer tracker will faithfully report that you are not mentioned and will not tell you that the cause is a robots rule. If your pages are perfectly readable and you still are not cited, a scanner that reports a clean bill of health is telling you the truth and not the thing you wanted to know. Seeing what a crawler receives and seeing what the models say are separate purchases. The split was set out in more detail in an earlier comparison of the two machines, and nothing in this run contradicts it.

What seven AEO pricing pages returned to a crawler

Each page was requested once, then once again to confirm the answer was stable rather than a one off. Six of the seven returned an HTTP 200 with a readable page both times, and the byte counts were identical across the two attempts in every case, so nothing below rests on a single sample.

The exception is Otterly.AI, whose pricing page answered HTTP 202 on both attempts with bodies of 179 and 176 bytes containing no words at all. That is a bot challenge rather than a page. It is a defensible thing for a company to do and it is not evidence of anything about the product, but it does mean that no figure from that vendor could be read and none is quoted here. A reader who wants those numbers has to fetch otterly.ai/pricing in a browser.

Three of the six readable pages carried no numeric plan price anywhere in their bytes. Profound publishes two columns, a free Trial described as being for companies who want to trial Profound before getting a demo, and an Enterprise column priced as Custom. The only currency figure in the page is the 180 million dollar Series D in its announcement banner, which is not a price. Peec AI publishes the contents of four plans in full, Starter, Pro, Advanced and Enterprise, without publishing what any of them costs in the returned document. The Semrush AI Visibility Toolkit pricing page returned 1,054 words and no price either.

None of that means the prices do not exist. It means they are not in the bytes a crawler is handed, which is a different and more specific claim, and it is the only one this test can support. Anyone reading those pages in a browser may well see numbers, and the three vendors that do publish in the bytes show that nothing about the category prevents it. Our method is deliberately the crawler's view and not the reader's.

Every tool meters a different unit

Ranking these products on one number is the standard way to write this post and it cannot be done honestly, because the number is not the same number. Four of the pages in this sample publish their allowance in four different units, all read on 24 September 2026.

PromptAlpha meters prompts tracked, and publishes the arithmetic plainly: a Starter plan at 107 dollars a month billed yearly with 3 answer engines tracked and 50 prompts tracked, and a Growth plan at 249 dollars a month billed yearly with 5 answer engines tracked, named as ChatGPT, Gemini, Perplexity, Grok and Claude, and 150 prompts tracked. AthenaHQ meters AI responses and says so in one line on its page: 1 credit = 1 AI response, with its free tier stated as including 300 credits. Those two units are close but not equal, because a prompt sent to five engines is one prompt and five responses.

Ahrefs meters checks, and its Brand Radar page publishes the conversion in its own worked example: each prompt is checked daily on every selected platform, 5 prompts on 1 platform equals 150 checks a month, and Claude consumes 8 checks per update. That last clause is the useful one, because it prices one model differently from the others inside a unit that looks flat. Peec AI publishes no price but does publish the shape of its meter, stating that usage is calculated by multiplying tracked prompts by active models and tracking frequency.

Four units, four different things a plan buys. A prompt count means nothing until you know how many models it is multiplied by and how often, and the honest way to compare two plans is to reduce both to responses per month and then check whether the engines are even the same ones. That arithmetic also explains why these products cost what they do: every unit is a paid model call, and unlike a page fetch it cannot be made cheaper by caching. It is also why a single week of tracking moves so much unless the frequency is high enough to average the noise out.

How many engines each tool says it covers

Engine coverage is the second number buyers compare, and it is published more consistently than price, though it is quoted at the top of a plan ladder rather than at the bottom. Profound's page states up to 9 Answer Engines tracked on its Enterprise column, and names ChatGPT, Gemini and Google AI Overviews on the free Trial. Peec AI's Enterprise column states up to 13 LLM models tracked, with the three lower plans each stating a choice of 3 models. PromptAlpha's ladder runs 3 engines on Starter, 5 on Growth and 10 on Enterprise.

Read those as ceilings, not as coverage. A count of engines says nothing about whether each one searches the live web at the moment it answers, which is the property that decides whether a page published last week can be cited at all. Lantad's own audit of the engines it can reach found that several of them do not search, and an engine answering from training data will happily name brands without ever fetching anything, so its answer is a fact about the model and not about your site. That audit is published as three of eight engines cannot search, and it is the reason the raw engine count is the weakest of the numbers on these pages.

The second thing a ceiling hides is the multiplier. Thirteen models on a plan that includes 350 prompts is 4,550 responses per cycle if every prompt runs on every model, and none of these pages promises that. Peec's own formula, prompts times active models times frequency, is the clearest statement of the trap: choosing more models with a fixed prompt budget either costs more or spreads the same budget thinner. When you evaluate a plan, fix the prompt list first, then ask what it costs to run that exact list on the engines you actually care about, which for most businesses is a short list. The same discipline applies to choosing the prompts themselves, where a vague prompt produces a moving answer no tool can stabilise.

Where Lantad sits, and where a rival is the better buy

Lantad belongs in this list in one row and not at the top of it. It is a read-side scanner first: it fetches pages as each crawler token, evaluates robots rules per token, and reports what survived the fetch. It does run prompts, and its own configuration is the honest source for how many. PLAN_LIMITS in core/src/config.ts sets promptsPerRun to 10, 15 and 25 across its paid rows and paidEnginesPerRun to 3, 5 and 5, with promptRerunsPerWeek at 1 on the lower rows and 2 on the higher. Those are settings somebody chose, not findings, and the live figures are on the pricing page.

Put next to the ladders above, that is a small answer tracker. If the question you are actually asking is what ChatGPT and Gemini said about your category this week, across a long prompt list, at daily frequency, with competitor share of voice over time, then Peec AI, Profound and PromptAlpha do that job at a breadth Lantad does not reach, and one of them is the right purchase. A buyer whose brand already appears in AI answers and who needs to watch the trend is buying an answer tracker, and should buy the one whose engine list matches the engines their customers use.

The case for the other machine is narrower and it is about order of operations. A tracker cannot report a cause, and the causes are unglamorous and common: pages whose text only exists after JavaScript runs, a robots rule written years ago that no longer means what its author intended, markup that a parser drops. Lantad's own scan of home pages loaded twice, once with scripting and once without, found the JavaScript-only share of prose was small in aggregate and total on a minority of pages, which is exactly the shape that makes client side rendering a per site question rather than a general one. If your pages fail prose parity, tracking answers first buys you a year of accurate reports about a problem the reports cannot name.

How to choose an AEO tool without booking a demo

The test run for this post is one a buyer can run in a minute, and it is worth running before any call, because it separates what a vendor publishes from what a vendor will tell you once you are in a pipeline. Fetch the pricing page and look for the number. If it is not there, that is not a scandal, but it does tell you the price is negotiated, which means it is a function of what the seller thinks you will pay.

Then apply four criteria, in this order, and state them before looking at any product so the ranking is not reverse engineered from whoever wins. First, which question do you need answered: what models say, or whether your pages can be read. Second, what unit does the plan meter, reduced to responses per month on the engines you care about. Third, how many of those engines actually search the live web rather than answering from training data. Fourth, what you can see without a sales call, since a published number is a commitment and a quoted one is an opening position.

Two of these are free to check on your own site before you spend anything. Whether AI crawlers are permitted is a robots question you can settle with a robots.txt tester in the time it takes to paste a hostname, and whether they arrive at all is visible in your own logs. If you sit behind Cloudflare, its AI Crawl Control documentation describes per crawler controls and request visibility available on all plans, which is a real part of this category being given away. Spending money on a tracker before those two checks is how a team ends up paying monthly to watch a number that a single robots line is holding down.

None of this ranks the field, and that is deliberate. On the evidence gathered here, three of six pricing pages did not publish a price to a crawler, one vendor did not return a page at all, and four vendors meter four different units. Those are facts about documents on one date, they will age, and the right response to any of them is to open the page yourself rather than to trust a list.