How do I compare my SEO and AI visibility with competitors?

Measured on one axis, most competitive positions look settled. Measured on both, they frequently are not — and the surprises run consistently in one direction.

Two axes, four quadrants

Plot each competitor on two dimensions.

Horizontal: organic strength. How they perform across your priority queries — positions, coverage, and roughly what backs them.

Vertical: AI presence. How often assistants name and cite them across a fixed set of buying questions.

Four quadrants result, and each implies a different threat and a different response. The value of the exercise is that most teams have only ever measured the horizontal axis, and the vertical one frequently contradicts it.

Choosing the competitors

Two lists, and they overlap less than expected.

Search competitors are whoever ranks for the queries you want. For most businesses that includes review sites, marketplaces, publishers and forum threads — none of them commercial rivals, all of them taking traffic. Analysing only your commercial competitors means studying sites that may not rank while ignoring the ones beating you.

AI competitors are whoever assistants name in your category. This list is discoverable in twenty minutes and routinely contains a surprise: a company you had not considered a rival, or a smaller player named consistently because three review sites describe them clearly.

Take the union, cap it at five, and favour anyone appearing on both lists.

Building the comparison

The SEO axis. Build a table with your priority queries down the side and you plus each competitor across the top, with current position in each cell. An hour of work, and it usually reorders priorities on its own. Look for the pattern rather than the worst cell — narrow losses everywhere is a different situation from one competitor dominating a segment.

The AI axis. Take twenty buying questions phrased the way a real buyer asks, run each against ChatGPT, Gemini and Perplexity at least twice in clean sessions, and record for every answer: which brands were named, in what order, and which URLs were cited.

Then compute, per competitor: mention rate across the prompt set, citation rate, and average position within the answers where they appear. Being named first in an answer is materially different from being named third.

Store the raw answers. Generated responses vary between identical runs, so a comparison built on single runs is comparing samples of randomness rather than positions.

What each quadrant means

Strong SEO, strong AI. The genuine market leader in visibility terms. Do not benchmark against them directly — benchmark against whoever sits immediately above you. Against this competitor the productive move is an asymmetry: going narrower, or publishing something they cannot replicate by writing harder.

Strong SEO, weak AI. The most common quadrant, and where most established companies sit. It usually means a mechanical failure — content that only renders after JavaScript, or blocked crawlers — that nobody has noticed because rankings are unaffected. If this is you, the fix is days of engineering work and it is the largest single improvement available. If this is a competitor, it is an opening.

Weak SEO, strong AI. The quadrant that surprises people. A company ranking below you being named ahead of you almost always has stronger third-party corroboration — described consistently across review sites, comparison posts and forums that assistants weight heavily. They are winning the shortlist without winning the results page.

Weak on both. Not currently a threat, and worth checking periodically rather than continuously.

The consistent surprise

Across this analysis, one finding recurs: the correlation between organic position and AI presence is weaker than people expect.

The reason is mechanical rather than mysterious. Assistants weight consistent independent description very heavily, and that is a different input from links and rankings. A company with modest SEO and a dozen accurate third-party descriptions can outperform a company with strong SEO and none.

Your Google position tells you less about your AI presence than almost anyone assumes. Measure both or you are guessing about half your visibility.

The second recurring finding is that a large share of cited sources are not company sites at all — they are review platforms, comparison articles, industry roundups and forum threads. That means some of your competitive position is being decided on pages you do not control and may never have read.

The citation map

The most actionable output of the whole exercise, and the one most often discarded.

Collect every URL cited across all the answers, ranked by how often it appears. That list is a map of what influences your category, derived from evidence rather than assumed.

Three things to do with it. Check where you appear and whether the description is accurate and current. Check where you are absent — those are your highest-value corroboration targets, far more valuable than generic coverage. And check what those pages have in common, because the pattern usually tells you which format your category’s answers are assembled from.

This list is worth more than the quadrant chart, because it converts a diagnosis into a specific brief with named targets.

The cheapest version of this analysis

The full two-axis comparison is a day of work. A useful eighty percent of it takes ninety minutes.

Pick your five most commercially important buying questions. Run each against ChatGPT and Perplexity twice, in clean sessions. Write down every brand named and every URL cited. Then check your own Search Console position for the closest equivalent query.

That produces the single most valuable output of the whole exercise: a short list of competitors named ahead of you, and a short list of sources cited instead of your pages. Both are directly actionable, and neither requires a tool, a budget or a project plan.

Do the full version quarterly. Do this version whenever someone asks how you are positioned, because the answer takes an afternoon rather than a fortnight and it is evidence rather than opinion.

Tracking it over time

A one-off comparison is a snapshot. The version that changes decisions is the trend, and it costs little more once the prompt set and query list are fixed.

Re-run the AI half monthly and the full analysis quarterly. Plot three lines: your mention rate and the mention rates of your two closest competitors, across the same fixed prompt set. Three lines on one chart answer every question anyone will ask — are we improving, are we gaining on them, and is it consistent across engines.

Two disciplines keep the trend meaningful. Keep the prompt set frozen, because every edit resets comparability and the temptation to change it arrives exactly when results are disappointing. And keep the competitor set frozen too, adding new entrants as an additional series rather than swapping them in, so the history remains readable.

Comparing against sources that are not companies

One complication most competitor frameworks do not handle: a large share of what beats you in AI answers is not a competitor at all.

Review platforms, comparison articles, industry roundups and forum threads appear repeatedly in citation lists. They are not taking your customers and they are absorbing the citation you wanted, which makes them a different kind of problem requiring a different response.

You do not compete with them — you get into them. Being accurately represented on a frequently-cited roundup contributes to every answer that cites it, which is more efficient than trying to outrank it. Track these separately from commercial competitors in your comparison, because conflating the two produces a strategy that fights the wrong opponent.

What movement actually looks like

Expect slow, uneven change. Competitive positions in AI answers shift over quarters rather than weeks, and short-term movement is dominated by the variance inherent in generated answers.

A meaningful gain looks like your mention rate rising from two-in-twenty to six-in-twenty over two quarters while a competitor’s holds steady. It does not look like a dramatic single-month jump, and when that appears it is usually sampling rather than progress — which is why running each prompt several times matters as much as running it regularly.

Acting on it

The analysis is finished when you can write four sentences.

1. Who actually competes with us, on each axis — usually a different list from the one you started with.
2. Which quadrant we are in, and therefore whether our gap is mechanical, editorial or external.
3. Which specific third-party sources describe competitors and not us.
4. Which terms and prompts are winnable this year, with the rest explicitly parked rather than silently carried.

If your analysis cannot produce those four sentences it was a description. If it can, it is a roadmap — and the quarter after it usually writes itself, because a named gap with a named target list is considerably easier to schedule than a strategy.

Sources

• Vercel and MERJ — analysis of 500M+ GPTBot fetches finding zero JavaScript execution; same behaviour for ClaudeBot and PerplexityBot.
• Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande — “GEO: Generative Engine Optimization”, ACM SIGKDD 2024 (arXiv:2311.09735).
• Google Search Console — first-party query, impression and position data used for the organic axis.

Frequently asked questions