Tracking Competitors in AI Search

To track competitors in AI search, pick three to six rivals, run one neutral prompt set across the engines your buyers use, and log every brand mentioned, its position, its framing and the sources cited. Repeat monthly on the same prompts. Our view: the most useful output is a gap list of prompts and sources where rivals appear and you don't.

Points marked "Our view" are interpretation; examples are illustrative.

On this page

• Step 1: Which competitors should you track?
• Step 2: How do you build a shared prompt set?
• Step 3: Which engines and how many runs?
• Step 4: What should you log?
• Step 5: How do you calculate competitor metrics?
• Step 6: How do you build a gap report?
• Step 7: How do you report it?
• What should you not do?
• How do you spot new competitors early?
• How do you act on what you find?
• How often should you refresh the competitor list?
• How do you keep the comparison fair?
• Exercise
• Checklist
• From MentionX data: where competitors show up
• Worked example: Ledgerline maps its rivals
• How do you choose prompts that test competitors fairly?
• How do you read competitor citations?
• How do you present competitor data without overreacting?
• Who should see the competitor report?
• What do you do when a competitor is described wrongly?
• How do you log runs so they can be compared later?
• Common questions
• Related
• Sources

Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.

Step 1: Which competitors should you track?

Start with the brands buyers compare you with: ask sales, check lost-deal notes and look at review site category pages. Then run ten category prompts and add any brand that AI answers name often but you hadn't listed. Our view: three to six is manageable. Track "others" as a single bucket.

Step 2: How do you build a shared prompt set?

Use the same prompts for every brand so comparisons are fair.

Our view: keep comparison prompts balanced, including ones that name rivals against each other, not only against you. See building a prompt set and prompt research.

Step 3: Which engines and how many runs?

Choose the engines your buyers use, for example ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews. Run each prompt at least three times per engine, because answers vary. Record the date, engine, mode and location.

Step 4: What should you log?

For each answer:

1. Every brand mentioned, in order.
2. Framing per brand: positive, neutral, mixed, negative.
3. Final recommendation, if any.
4. Cited sources (URLs and domains).
5. Any incorrect facts about any brand.

A spreadsheet works for a few hundred answers. Beyond that, a tracker saves time; see manual vs automated tracking.

Step 5: How do you calculate competitor metrics?

• Mention rate: share of answers naming each brand.
• Share of voice: each brand's mentions divided by all tracked brand mentions; see measuring AI share of voice.
• Average position when mentioned.
• Recommendation rate: share of answers where the brand is the final pick.
• Source overlap: domains cited for rivals but not for you.

Step 6: How do you build a gap report?

Filter to answers where a rival is mentioned and you aren't. Group by prompt type and by cited domain. Our view: the top ten domains in that list are your earned-media targets, and the prompt groups show which content or positioning is missing.

Step 7: How do you report it?

Show a monthly table of mention rate and share of voice per brand per engine, with sample sizes. Add the top five gaps and one action each. Flag changes only when they're outside normal variation; see statistical confidence.

What should you not do?

• Don't contact or post about competitors misleadingly; that risks platform rules and law.
• Don't run prompts only from branded angles.
• Don't change the prompt set every month.
• Don't treat one engine as the whole market.

How do you spot new competitors early?

Our view: review the "others" bucket monthly. When a brand that wasn't on your list appears in several answers across engines, add it to tracking. New entrants often show up first on problem prompts and in community sources before they reach "best" lists, so watch those prompt groups closely.

How do you act on what you find?

Our view: pick no more than three gaps per quarter. For each, write the hypothesis ("Beta is named on chasing-invoice prompts because of its how-to guide on X"), the action (publish a better guide, update a review profile, pitch a list), and the prompts you'll recheck. Recheck after a full month so engines have time to recrawl. Record outcomes, including no change, so your team learns which actions work.

How often should you refresh the competitor list?

Our view: quarterly. Ask sales which names come up in calls, check review site categories again and look at the "others" bucket. Remove brands that no longer appear and add any that have grown, noting the change in the report so trends remain comparable.

How do you keep the comparison fair?

Our view: run all brands through exactly the same prompts, engines, dates and locations, and use the same labelling rules for framing. Avoid logged-in accounts with history that might personalise answers. If you change anything, such as adding an engine, start a new baseline rather than mixing periods. Note any known events, such as a rival's product launch or outage, alongside the numbers so readers don't misread a spike as a trend.

Exercise

Pick three competitors. Run ten neutral category prompts three times each in two engines. Calculate mention rate per brand and list the five domains cited most often when a rival appears and you don't.

Checklist

• Three to six competitors chosen with sales input
• Neutral, shared prompt set
• Multiple runs per prompt per engine
• Brands, position, framing and sources logged
• Gap report by prompt type and domain
• Monthly report with sample sizes

From MentionX data: where competitors show up

Method. Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

Where competitors appear depends heavily on the type of prompt. In our sample, brands were named in almost every comparison answer but rarely in problem-led answers:

Engines also differ in how often they name any brand:

Our view: split competitor tracking by intent. Comparison prompts show head-to-head standing; best-of and recommendation prompts show who gets shortlisted. Problem prompts show who the engines treat as the default answer, and that is where new entrants often appear first.

Limits. A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

Worked example: Ledgerline maps its rivals

Ledgerline is a fictional accounts-payable tool. Figures are illustrative.

Ledgerline tracks four rivals, Competitor A to Competitor D, on 80 prompts across three engines, with three runs each. It logs every brand named and every cited URL.

Competitor A leads on best-of prompts in two engines. Competitor B is rarely named but is cited often, because its documentation answers detailed how-to questions. Competitor C appears mostly in one engine. In problem prompts, a fifth brand Ledgerline wasn't tracking shows up repeatedly. The team adds it as Competitor D for the next quarter and drops a rival that never appears. The gap report shows 14 prompts where Competitor A is named and Ledgerline isn't; nine of them cite the same two comparison articles. The team plans outreach to both publishers and a new comparison page.

How do you choose prompts that test competitors fairly?

Use prompts a buyer would actually type, not prompts that name your brand. Include unbranded best-of questions, comparisons between two rivals, comparisons that include you, and problem questions with no brand at all. Keep the same wording for every brand so nothing is tilted in your favour. If you write separate prompts for each competitor, write them from the same template.

Avoid loading prompts with your positioning. A prompt such as "best invoice tool for teams that hate manual data entry" may favour you if that's your tagline. Ask a colleague outside marketing to review the list for bias.

How do you read competitor citations?

Citations show which pages shape an answer. Group them by domain and by page type: competitor-owned pages, review sites, publishers, forums and documentation. Where a rival's own pages are cited, study what those pages answer that yours don't. Where third-party pages are cited, check whether you appear on them at all.

The URL mix in our sample shows how spread out citations are:

Our view: blog and article pages take the largest share, so independent articles that compare tools are usually the first place to look when a competitor leads.

How do you present competitor data without overreacting?

AI answers vary from run to run. A competitor gaining a few points in one week is often noise. Show ranges, not single values, and highlight only changes that persist across two or more measurement rounds. When a rival's share jumps in every engine at once, look for a cause: a launch, funding news, a new review round-up or a big content release.

Who should see the competitor report?

Product marketing uses it for positioning. Content uses the gaps to plan pages. PR and partnerships use the cited-source list for outreach. Sales may want the comparison answers for objection handling. Send each group the part it can act on, not the whole spreadsheet.

What do you do when a competitor is described wrongly?

Leave it alone. Correcting rivals isn't your job, and claims about them can create legal risk. Focus on making sure engines describe your own product accurately and that your comparison pages are fair and sourced.

How do you log runs so they can be compared later?

Use one row per brand per answer. Record the date, engine, prompt, prompt intent, run number, whether the brand was named, its position in the answer and the URLs cited. Keep the raw answer text in a separate column or file. When someone asks later why a rival jumped, you can reread the answers instead of guessing. Store the prompt set with a version number, so you can tell whether a change in results came from the engines or from edits to your prompts.

Common questions

How many competitors should I track? Three to five named rivals is enough for most teams, plus a check for new names.

Should I track competitors' branded prompts? A few, to see what engines say about them, but focus on unbranded buyer prompts.

Can I compare scores from different tools? Not reliably. Tools use different prompts, engines and formulas.

Related

• Measuring AI share of voice
• Run an AI visibility audit

Sources

This tutorial describes a method rather than citing platform rules. For how Google's AI features select supporting links, see Google, AI features and your website. All accessed 30 Oct 2026.