There is a company in your category that assistants name when yours goes unmentioned. That is annoying, and it is also the most useful research asset available to you, because whatever signals persuaded a model to recommend them are almost entirely visible from the outside.
Conventional competitor analysis does not surface them, since it looks at rankings and backlinks rather than at what makes a company legible to a model. The process below is what any serious approach to AI Search Optimization Services starts with, and you can run most of it yourself in a day.
Step One: Get the Model to Explain Its Own Reasoning
Start by asking the assistant why.
Run your main category question, note who gets named, then ask what makes that company a good fit for the request. The response is not authoritative and should not be treated as such, but it does reveal which attributes the model has attached to that competitor and considers relevant. Certifications, service scope, geographic coverage, specialization, years in operation, particular capabilities.
Write those attributes down. That list is your competitor’s effective positioning as far as the model is concerned, which is often noticeably different from the positioning on their own website.
Then ask the same question about your company. The gap between the two attribute lists is your actual worklist, and it is usually shorter and more specific than a strategy document would suggest.
Step Two: Find Where Their Attributes Are Confirmed
Now trace each attribute back to its source.
Search each claim alongside their company name and see what comes up. You are looking for where the confirmation lives, not whether it is true. Typically you will find some combination of:
- Their own site, stating it plainly and in an extractable position
- Directory or association listings carrying the same detail
- Third party coverage or partner pages repeating it
- Structured data making the attribute machine readable
The pattern that matters is how many independent sources say the same thing. An attribute confirmed in four places behaves very differently from one that appears only on their homepage, and that difference is frequently the whole explanation for why they get named and you do not.
Step Three: Compare Extractability, Not Quality
Open their most important commercial page next to yours and read both as a machine would.
Specifically, check where the concrete facts sit. Many winning pages are worse written than their competitors’. What they do differently is state things flatly and early: what the company does, for whom, where, with what scope, at what stage of the process. No narrative preamble, no adjectives standing in for specifics.
A useful test is to lift a single sentence out of each page at random and ask whether it still makes sense with no surrounding context. Pages that pass that test get extracted. Pages that need their own paragraph to make sense do not.
Ken Ganley Mentor CDJR, an Ohio dealership selling Chrysler, Dodge, Jeep, and Ram, illustrates what closing this specific gap does. The dealership was a substantial local business with almost no page level presence in AI answers, and the work centered on making inventory and service detail concrete and extractable rather than on building anything new. The movement came inside two months.
Step Four: Map Their Question Coverage Against Yours
Buyers ask a sequence, not a single question, and the earlier questions shape everything after.
List the questions somebody works through before they are ready to choose a provider in your category. Then check each one to see who appears. What you frequently find is that a competitor owns the early orientation questions, which lets them frame the criteria by the time the comparison question arrives.
That is a more valuable position than winning the final question, and it is usually less contested.
Note also which questions nobody wins. Those are the cheapest openings available, and they tend to be the specific, awkward, low glamour questions that no marketing department wanted to build a page around.
Step Five: Check Whether Their Advantage Is Structural or Temporary
Some competitor positions are defended. Many are not.
Look for signs of active maintenance. Is their content dated and refreshed? Do their external listings look current or abandoned? Is their presence consistent when you repeat the same query across a fortnight, or does it fluctuate?
Fluctuating presence means they are not defending, which means the position is available to anyone who does. Consistent presence across repeated tests suggests somebody is working on it, and displacing them will take sustained effort rather than a single push.
This distinction determines whether you are planning a two month project or a four quarter programme, and getting it wrong in either direction is expensive.
Step Six: Separate What You Can Copy From What You Cannot
Be honest at this stage, because it saves a quarter.
Usually copyable: extractable page structure, schema coverage, directory accuracy, question coverage, clarity of service description, geographic signals.
Sometimes copyable: third party coverage, association memberships, partner relationships. Achievable, but on a longer timeline and often through routes that need real relationships.
Not copyable: genuine credentials you do not hold, operating history you do not have, scale you cannot claim. If a competitor’s advantage rests substantially on these, contest a different question set rather than the same one.
That last case is more common than people admit, and recognizing it early is a strategic win rather than a defeat. Adjacent questions where their credential is irrelevant are usually wide open.
Turning the Analysis Into a Sequence
The output of the exercise above should be four short lists, in this order of action.
- Attributes they have confirmed and you have not, where the claim is true for you too. This is the fastest work: state it clearly, get it confirmed externally.
- Pages of yours that fail the extraction test. Editing, not writing, and it moves quickly.
- Questions nobody currently wins. Your cheapest new content, and the least contested.
- Questions they own through genuine advantage. Deprioritize or approach obliquely.
Most companies find that lists one and two account for the majority of the gap, which is why the diagnosis is worth doing before commissioning anything larger.
EC Council, the global certification body behind CEH and CND, is a reminder that scale changes the timeline rather than the sequence. Already globally recognized, the work still took six months of technical restructuring across a large content estate. The order of operations was the same. The volume was not.
Where This Analysis Reaches Its Limits
Two things you cannot see from outside, worth knowing before you over interpret.
You cannot see how much of their position came from deliberate work versus historical accident. A company with fifteen years of accumulated coverage may not have optimized anything. That matters, because copying their current state does not replicate the accumulation that produced it.
You also cannot see their trajectory from a single snapshot. Repeated testing over several weeks solves this, and it is the difference between a real read and a lucky one. Independent measurement matters here too, since dashboards from any vendor reflect their own weighting, and reviewing what the current best chatgpt seo software options actually measure is worth doing before treating any single figure as settled.
Pick the one competitor that appears most often where you do not, and run steps one and two on them this week. The attribute gap that comes back is usually the shortest useful worklist anyone in your position will get.