Ranking and recommendation use different evidence. Google ranks a page for a query; an engine names a company when multiple independent sources describe it the same way. One page cannot supply that.
So position one and total absence are not a contradiction. They are two systems answering two different questions, and doing well at the first tells you almost nothing about the second.
This is the single most disorienting result in AI visibility, and it arrives in the same shape every time. Somebody checks the rankings, sees position one for the money keyword, then asks ChatGPT the same thing in plain English and finds four competitors and no mention of themselves.
Nothing is broken. The two systems are being asked different questions, and the answer to one is not evidence for the other.
What is each system actually being asked?
A page-level judgement. One excellent page, well linked and well structured, can win it outright. Your page is the unit competing.
An entity-level judgement, assembled from many documents. Your page is one input among several, and most of the others are not yours.
A page can be the best answer to “what is field service scheduling software” and contribute almost nothing to “what are the best field service scheduling tools”, because the second question is not asking which page is good. It is asking which companies to name, and companies are established across sources rather than on one.
Why does your best page not carry the weight?
Because of what it is, structurally, rather than how good it is.
Every vendor in your category has a page claiming to be the right choice. To a retriever assembling a comparison answer, forty vendor pages are forty instances of the same genre, all self-interested and all equally available. There is no signal in preferring one of them, so the useful material comes from documents where the author is not the subject.
This is not a value judgement about your writing. Your page is doing its job: it is almost certainly the authoritative source for what you are. It just cannot be the authoritative source for whether you belong on a shortlist, and no rewrite changes that.
Does ranking help at all?
Yes, and it helps most on Google’s own surfaces, where the connection is direct and documented.
For AI Overviews and AI Mode, a page qualifies as a supporting link if it is “indexed and eligible to be shown in Google Search with a snippet”, and Google states there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”. Your ranking work is the entry ticket there, and losing it would cost you both.
On the assistant surfaces the relationship is looser. OpenAI and Perplexity run their own crawlers over their own indexes, so your Google position is not an input to their retrieval at all. Being good enough to rank correlates with being good enough to retrieve, but it is a correlation rather than a mechanism.
Direct. Indexed and snippet-eligible is the stated requirement, and that is exactly what your SEO work produces.
Strong linkIndirect. Separate crawlers and separate indexes, but the same underlying qualities tend to make a page retrievable in both.
Correlation onlyAlmost none. Being named is decided by how the wider corpus describes you, not by where any single page sits.
No linkWhy does this feel like a contradiction?
Because a decade of practice trained everyone to treat ranking as a proxy for visibility, and it was a good proxy right up until the format changed.
When results were ten links, position was visibility. Rank first and you were seen; rank eleventh and you were not. The proxy held so reliably that nobody had to distinguish between “our page is the best answer” and “our company is one of the options”, because on a page of links those collapsed into the same outcome.
An answer separates them again. It names companies and cites sources, and those are two different lists. You can be in the second and not the first, which is the specific and disorienting case of being cited as a source for a claim about your category while not being named as one of the options in it.
Best case. The engine recommends you and points at you as a source.
Your excellent page taught the engine about the category. It then recommended four competitors using what it learned.
Somebody else’s page said you were good, and the engine believed them rather than you.
How do you confirm this is what is happening?
Two checks, ten minutes, and they are conclusive together.
In a clean, memoryless session, ask what your company is and who it is for. Read whether the description is specific and correct.
Accurate here proves retrieval is working. The problem is downstream.Same session is fine. Ask what the best tools in your category are for a buyer like yours, three times, and record which brands appear.
Accurate in step 1 and absent here is the confirmed pattern.Collect every source the step 2 answers cited. Note how many are review platforms, comparison articles and community threads rather than vendor sites.
That list is where the recommendation is actually being decided.So what do you do about it?
Work on the layer the second question actually reads, which is not your website.
Review platforms, directory listings, independent comparison pages and genuine community presence are what supply comparative evidence. It is slower than publishing and it is the thing that moves the number. The mechanics of that are in why your SaaS doesn’t appear in AI recommendations, and the wider question of which of the four causes you have is in the diagnostic guide.
One thing not to do: stop the SEO. On Google’s surfaces it is the entry requirement, and dismantling it to fund AEO work removes you from the results you currently win as well as the ones you do not. The shift is in proportion, not in direction, and the fuller version of that argument is in AEO vs SEO.
How long before the second system catches up?
A quarter, and it is worth setting that expectation before anyone asks for a weekly update.
The delay is structural rather than arbitrary. Comparative evidence has to be published, then crawled, then retrieved for a relevant question, then chosen by a generator as worth summarising. Each step has its own lag and none of them accelerate because you are impatient. A team promising a six-week turnaround on recommendation-list absence is promising something the mechanism cannot deliver.
What does move inside six weeks is the input side: more independent domains mentioning you, and more of the sources that recur in your category carrying your name. Report that as a leading indicator, honestly labelled, and it is enough to keep the work funded while the output catches up.
Ranking first and being absent is not a paradox to resolve. It is a measurement telling you that you have solved one problem completely and never worked on the other one.