Can you actually influence AI answers?
You have been told your company needs to show up when a homeowner asks ChatGPT who to call. You have probably been quoted a monthly figure for it. What nobody sells you is a way to tell whether any of it worked.
This page sorts the whole question into three piles: what you can check yourself today, what merely repeats often enough to act on, and what nobody outside these companies can tell you at all. The sorting is the argument. Most of what gets sold as answer engine optimisation sits in the second and third piles while being priced like the first.
What you can check yourself, today
Verifiable · each item names its own test
Whether an engine can retrieve your pages at all
GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended. Your robots.txt either allows them or it does not, and you can read the file in a browser right now. This is the only lever on this page whose mechanism nobody disputes: a blocked crawler removes you from the retrieval pool for the part of these systems that reads the live web.
Whether your business facts are machine-readable and agree with each other
Schema validates or it throws errors. Your name, address, phone and service area either match across your site, your Google profile and the directories, or they do not. An afternoon's work tells you which, and it needs no vendor.
What the engines say about you right now
Write down a fixed set of questions homeowners actually ask. Run them, logged out, from your market, and record the answers word for word with the model and the date against each one. That is a measurement, and another person can repeat it and get a comparable result. This is the part most often described rather than done.
Whether a given search even produces an AI answer
Plenty of commercial searches still do not, and presence shifts by phrasing, location and device. Check the exact search you care about before paying anyone to optimise for a feature that may not appear on it.
What repeats, without a mechanism anyone has shown
Correlation only · observed pattern, no proven cause
Everything in this section is real and worth knowing. None of it has been shown to cause anything.
Cited pages tend to answer the question near the top
Look at the pages an AI answer links and you will usually find the answer stated plainly in the first few lines rather than buried under a long introduction. That pattern holds often enough to write your pages that way. What nobody has published is a controlled result showing that restructuring a page caused a citation.
Businesses with more reviews get named more often
Also true, and also confounded. Review count travels with being older, larger and better known, and any of those could be doing the work. Nobody has separated them.
Reddit threads, listicles and comparison articles show up in source lists
Measurable, and several agencies now sell placement into exactly those sources. Whether seeding them earns the citation, or whether both follow from already being a known business, is unestablished. There is a second question underneath the first, which is whether you want your company placed into third-party discussions that way. Decide that one knowingly rather than finding out from an invoice.
What nobody can tell you from outside
Not knowable · no method exists to settle these
Why a model named one company and not another
There is no rank, no log, no diagnostic. Ask the same question twice in one afternoon and you can get two different sets of names. Any explanation of a single result is a story told after the fact.
Whether your change caused the change in output
You cannot run your market twice. There is no version of your town where you did not publish the page, so every before-and-after in this field is an observation, not a result. That includes ours.
Whether llms.txt does anything
No engine has confirmed reading it. It costs almost nothing to publish and there is no evidence it moves anything. Several companies we scored describe it as the file AI reads to decide who to recommend, which is a mechanism nobody has demonstrated.
How many of your calls came from an AI answer
Most of this traffic arrives with no referrer at all, or shows up later as someone searching your company name directly. Any dashboard giving you a confident figure is estimating and should say so.
Whether today's measurement survives the next model update
Often it does not. A prompt set that returned you consistently in March can stop doing so in April with nothing changed on your side.
What a defensible measurement looks like
The standard we score against · criterion 2 on the leaderboard
None of the above means measurement is impossible. It means the measurable thing is what the engines say, and the unmeasurable thing is why. Anyone selling you the second should be asked how they know. Here is the standard we score agencies against, and you can hold your own vendor to it whether or not they appear on this site.
- A fixed prompt set
- Written down before the work starts, so it cannot be quietly changed to a set you do better on.
- Named engines
- Named individually, with the model version where the engine exposes one. “AI” is not an engine.
- A stated location
- Logged out, from the market you actually serve. These answers are local and personalised.
- Several runs, reported as a rate
- Because the systems are non-deterministic, a single check is noise. Named in six runs out of ten is a finding; named once is an anecdote.
- Verbatim capture
- What each engine actually said, kept alongside any summary score, so you can read it rather than trust a number.
- A competitor set
- Measured the same way on the same day, because the interesting question is who got named instead of you.
- A failure condition
- Written down in advance: what result would make them tell you this has not worked. An agency that cannot answer this has not given you a way to fire them.
The short version
You can change the inputs and you can measure the outputs. Nobody can currently draw the line between the two for a single business, including us. The gap between those two things is where most of this industry makes its money, and the honest response is not to avoid the work but to insist on knowing which half you are buying.