Competitor keyword analysis for SEO, AEO, and generative search

Why does competitor keyword analysis look different now?

For years, competitor keyword research had one job: find the phrases your rivals rank for, then decide which ones you could realistically take. That job has not gone away. It just stopped being the whole picture.

Buyers now research in more places. Some still type a phrase into Google and scan ten blue links. Some read an AI Overview and never scroll. Some open ChatGPT or Perplexity and ask a full sentence question, then follow whichever two or three sources the answer cites.

That means a competitor can beat you in three different ways. They can outrank you, they can own the snippet or overview, or they can be the source an AI assistant repeats. A useful analysis checks all three.

The good news is the underlying work is familiar. Keyword research is still keyword research. You are just adding two more columns to the same spreadsheet.

Which competitors should you actually analyze?

Start with a short list. Five to eight names is plenty. More than that and the analysis turns into data collection instead of decision making.

I usually sort competitors into four buckets:

• Direct competitors. Same services, same buyer, same region.
• Aspirational competitors. Bigger or better known, useful for seeing where the category is heading.
• Search competitors. Publishers, directories, review sites, and marketplaces that outrank everyone for your informational queries even though they do not sell what you sell.
• AI-cited competitors. Whoever shows up when you ask an assistant the questions your buyers ask.

That last bucket surprises people. A regional professional services firm I worked with assumed its competitive set was three local peers. When we asked five assistants for recommendations in that city, the same national comparison blog appeared in four of the five answers. It was not a business competitor at all, but it was shaping the shortlist.

Build the list before you touch a tool. Otherwise the tool builds it for you, and tool-generated competitor lists tend to favor whoever has the largest keyword footprint rather than whoever is winning your buyers.

What data do you collect for each competitor?

Keep the collection boring and repeatable. I run four passes.

Pass one: classic keyword overlap

Pull organic keywords, estimated traffic, and top landing pages for each competitor from a rank tracking tool. Then run a keyword gap report so you can see the terms two or three competitors rank for and you do not. Those shared gaps are usually the most reliable signal in the whole exercise, because multiple companies validating a term is stronger evidence than one.

Add your own Search Console data next to it. Terms where you already earn impressions but sit in position eight through twenty are almost always faster wins than brand new topics.

Pass two: SERP feature ownership

For your priority terms, note what the results page actually looks like. Is there an AI Overview? A featured snippet? A People Also Ask block? A local pack? A video carousel?

Then note who owns those features. A competitor holding the snippet for a term you both rank for is taking clicks you never see in a rank report.

Pass three: answer and citation testing

This is the pass most teams skip. Write twenty to forty real questions a buyer would ask, phrased conversationally, then run them through the assistants your audience uses. Record which brands are named, which URLs are cited, and how your business is described when it appears at all.

Do this in a clean session without personalization, and repeat it monthly. Answers vary, so you are looking for patterns rather than single results. My post on how to rank in ChatGPT and Perplexity covers the citation mechanics behind what you will see.

Pass four: content and structure teardown

For the top two or three pages that keep winning, look at how they are built. Word count matters less than shape. I note the heading pattern, whether answers appear in the first sentence under each heading, whether there is a comparison table, whether an FAQ block exists, and whether the page carries structured data.

How do you score a keyword across SEO, AEO, and generative search?

Once the data is in, score each keyword three ways instead of one. I use a simple three-column rating of high, medium, or low.

SEO opportunity. Is there real search volume, is the intent commercial enough to matter, and is the current top ten beatable given your authority?

AEO opportunity. Is the query phrased as a question, does the results page already show a snippet or overview, and is the answer short enough to state in two or three sentences? Definition queries, comparison queries, cost queries, and process queries score high here.

Generative opportunity. Do assistants answer this question with named brands or sources? Are those sources beatable? Is the topic one where a buyer asks for a recommendation rather than a fact?

A term that scores high on all three goes to the top of the roadmap. A term that scores high only on SEO is still worth doing, but you write it differently. A term that scores high only on generative search often belongs in a comparison or selection-criteria page rather than a service page.

This scoring is also how you avoid the trap of chasing volume. A query with 90 monthly searches that produces an AI recommendation for your category can be worth more than a 5,000 volume term that never converts.

How do you turn the analysis into a plan?

Map every winning keyword to one page. One primary topic per URL, supporting terms grouped underneath, and a clear note about whether the page is new, a rewrite, or a refresh. That is the same discipline I described in using AI to create keyword maps, and it holds up well when you add answer targets to the sheet.

My working columns look like this:

• Target URL and page type
• Primary keyword and supporting keywords
• Search intent
• Answer-target questions, written the way a person would say them
• Competitor pages currently winning
• SEO, AEO, and generative scores
• Structural requirements, such as a table, an FAQ block, or schema
• Internal links to add, in and out
• Owner, priority, and status

One practical example. A software services firm found that three competitors ranked for a pricing-model comparison query while assistants cited a single vendor blog for the same question. The plan was not a blog post. It was a dedicated comparison page with a real table, a two-sentence answer under each heading, an FAQ block, and internal links from the two service pages that supported it. Three months later the page held the snippet and started appearing as a cited source in about a third of test prompts.

How do you measure whether it worked?

Track four things, not one.

Rankings and impressions tell you whether the SEO side is moving. Search Console impressions usually shift before positions do.

Snippet and overview presence tells you whether the AEO structure is landing. Spot check your priority queries monthly and log what you see.

Citation frequency tells you whether generative visibility is improving. Keep the same prompt list every month and record how often you appear. Consistency of the prompt list matters more than the size of it.

Engagement and conversions tell you whether any of it mattered. AI-referred traffic tends to arrive in smaller volume with higher intent, which is one more reason engagement metrics beat traffic volume as a scorecard.

Review the whole analysis quarterly. Competitor sets change, assistants change how they cite, and a term that looked out of reach in the spring can be winnable by the fall.

What mistakes come up most often?

Copying instead of comparing. If you rebuild a competitor page one for one, the best you can do is tie. Look for the question they answered badly.

Analyzing only the competitors you already worry about. The publisher outranking everyone for your informational terms is shaping demand even if it never sends you a lead.

Treating AI citations as random. They vary, but they are not noise. Run the same prompts long enough and clear patterns show up.

Building a beautiful spreadsheet nobody uses. If the analysis does not end in named pages with owners and dates, it was research, not strategy. If you want help turning one into a build sequence, that is the core of my content strategy work.

Competitor keyword analysis FAQ

How many competitors should a keyword analysis include?

Five to eight is usually enough. Include direct competitors, one or two aspirational competitors, the publishers and directories that outrank everyone for your informational queries, and any source that AI assistants cite when asked about your category.

How do you find keyword gaps for AI search?

Standard keyword gap reports cover rankings. For AI search, build a list of twenty to forty buyer questions, run them through the assistants your audience uses, and record which brands and URLs get cited. Gaps appear where competitors are named consistently and you are not.

Should you target keywords or questions?

Both, mapped to the same page. Keywords define what the page is about and questions define the passages inside it. A service page can target a commercial keyword while answering four or five specific questions in question-style headings.

Does search volume still matter for AEO and generative search?

It matters less as a sole filter. A low-volume question that triggers an AI recommendation in your category can produce better leads than a high-volume informational term. Weigh commercial intent and answer opportunity alongside volume.

How often should competitor keyword analysis be refreshed?

Refresh the full analysis quarterly and rerun the AI citation prompt test monthly. Assistant answers shift more quickly than rankings, so the prompt test needs a shorter cycle to be useful.