2026-09-17, Analytics · Agencies · Fundamentals

Competitive analysis template with a data source for every row

A competitive analysis template is a fixed grid of questions you answer the same way for every competitor, with the source and date of each answer written next to it. The source column is what makes it useful: without it, a template mixes last year’s pricing, a salesperson’s impression and a ranking checked this morning, and nobody can tell which is which. Copy the tables below, keep one file per competitor plus a summary, and refresh each section on the cadence at the end.

Source labels used throughout:

• Their site: the competitor’s own pages (pricing, product, docs, changelog, press).
• Public pages: third-party pages such as review sites, app marketplaces, filings, job boards.
• API: this API’s endpoint and response field, collected on a schedule.
• Manual: research that needs a person: demos, trials, sales call notes, win/loss interviews.

0. Header

Fill in the domains and aliases row first. Every automated row matches on it, so a competitor whose docs live on a separate domain will look invisible in search unless that domain is listed.

1. Positioning

The last two rows show the gap between what a company says and what search engines and assistants repeat. When ChatGPT’s description of a rival lags their current positioning, start with the pages it cites (sources[].url) to see which outdated descriptions it is drawing on.

2. Pricing and packaging

Rule: if pricing is not published, write “Not published” rather than a remembered number. A price from a sales call goes in its own row with the date and deal context.

3. Product

Score only what you can link. A capability that exists “according to a prospect” is a lead to verify, not a 2.

4. Organic search presence

Fill from a fixed keyword set, tagged by group, for each market in the header. All rows below come from POST /v1/monitor/google (task type GOOGLE).

Not in this section: traffic, search volume and backlinks. The API does not return them. If you need them, add rows sourced from your analytics or a keyword tool, and label those rows as estimates.

5. AI answer presence

Two sources: the Google AI Overview on your keyword set, and assistant answers on a fixed prompt set.

entities[] exists only on ChatGPT; for other engines, match brand names and aliases in text. Sample each prompt several times per period, because answers vary between runs. AI share of voice has the full metric set.

6. News and announcements

News rows use the Google News endpoint.

Headlines are leads. Confirm material facts on the competitor’s own press page or filings before they go in a board summary.

7. Paid search and ads

Spend, bids and click data are not observable in any of these. Report presence rates, not “they spend more”.

Filling the automated rows

One batch per run covers sections 4, 5 and 7 for a keyword and prompt set. A minimal TypeScript submitter with fetch:

const API = "https://api.answerline.dev";
const headers = { Authorization: `Bearer ${process.env.API_KEY}`, "Content-Type": "application/json" };
const webhook = { url: "https://intel.example.com/hooks/template" };
const day = new Date().toISOString().slice(0, 10);

type Kw = { id: string; text: string; market: string; aio: boolean };
type Prompt = { id: string; text: string; market: string };

function tasks(keywords: Kw[], prompts: Prompt[], adSamples = 3) {
  const out: object[] = [];
  for (const k of keywords) {
    for (let s = 0; s < adSamples; s++) {
      out.push({
        taskType: "GOOGLE",
        idempotencyKey: `tpl:${k.id}:${k.market}:${day}:${s}`,
        webhook,
        payload: {
          query: k.text,
          country: k.market,
          ...(k.aio && s === 0 ? { include: { aioverview: { markdown: false } } } : {}),
        },
      });
    }
  }
  for (const p of prompts) {
    out.push({
      taskType: "CHATGPT",
      idempotencyKey: `tpl:${p.id}:${p.market}:${day}`,
      webhook,
      payload: { prompt: p.text, country: p.market },
    });
  }
  return out;
}

async function submit(all: object[]) {
  for (let i = 0; i < all.length; i += 500) {
    const res = await fetch(`${API}/v1/async/task/batch`, {
      method: "POST",
      headers,
      body: JSON.stringify(all.slice(i, i + 500)),
    });
    if (!res.ok) throw new Error(`batch ${i}: ${res.status}`);
    const body = await res.json();
    for (const r of body.results) if (!r.success) console.warn(r.index, r.error?.code);
  }
}

Sample 0 carries the AI Overview and feeds organic rows; samples 1 and 2 only add ad observations. Verify webhook signatures before storing results (how). Competitor SEO tracking covers the storage model and change detection.

Cost

Credits per async task: Google Search 3, plus 2 with the AI Overview; ChatGPT 5; Google News 3. For 100 keywords with 3 samples per run (AI Overview on the first) and 30 ChatGPT prompts, run weekly: (100 × 5 + 200 × 3 + 30 × 5) × 4 = 5,000 credits a month. Adding 10 Google News queries weekly adds 120. See pricing.

Scoring

Score each competitor per dimension on 0 to 3, with written criteria so two analysts arrive at the same number.

Threat score = Σ(score × weight) ÷ Σ(3 × weight), from 0 to 1. The thresholds in the search and AI rows are examples; set yours after the first full run so the scale spreads competitors out.

Rules that keep scores honest:

1. Every score links to the evidence row it came from.
2. Two people score independently; differences over one point get discussed, not averaged.
3. A score cannot change without a new evidence row and date.
4. Report the threat score with the per-dimension scores beside it. A single number hides why.

Refresh cadence

Monitoring cadence goes deeper on how often AI answers need sampling. For which tools fill the manual rows, see competitive intelligence tools by job, and for 18 ready questions with their exact fields, competitive intelligence examples.

Pitfalls

• No dates. A template without a “checked” column rots silently.
• Mixing estimates and observations. Traffic and volume figures are modelled; rankings, ads and citations are observed. Label them differently.
• One sample of volatile rows. Ads and AI answers vary between requests. Use rates.
• Missing aliases. Unlisted domains and product names make a competitor look absent.
• Scoring from memory. If there is no evidence link, the cell stays empty.

The metric definitions behind sections 4, 5 and 7 are on the competitor analysis page. To start collecting, see the quickstart.

Keep reading