Honest comparison structure beats affiliate sludge. Sections, tables, and the disclosure rules that keep trust intact.
Someone asks an assistant, "Northstar Analytics vs SheetStack for B2B funnel reporting — which should I pick?" The model reaches for a comparison page. If yours is honest, structured, and crawlable, you might be the source. If yours is affiliate sludge with fake scores, the assistant may prefer a third-party review instead — or invent a mashup of both.
Comparison pages assistants recommend from are boring in the best way: clear criteria, a table, who each option is for, and disclosures that keep trust intact.
Why comparison pages get retrieved
Answer engines like pages that already look like answers. A comparison with a decision table is easier to quote than a 3,000-word narrative. Similarweb’s Gen AI Landscape 2025 (US desktop, Sep 2025) reported ChatGPT referrals averaging about 15 minutes on site versus about 8 for Google, and roughly 7% conversion on transactional sites versus about 5% for Google — visitors who arrive from AI often already have intent. A clear comparison helps them finish the decision without feeling tricked.
You still cannot force the citation. You can make the page the kind of source retrieval systems prefer. Pew Research (March 2025) found that 88% of AI Overviews cited three or more sources — comparison pages with clear criteria are built to be one of those sources, not a vague "ultimate guide."
Structure that works
| Section | Purpose | Rule |
|---|---|---|
| One-sentence verdict framing | Sets scope | "Depends on X / Y" — not "we always win" |
| Criteria table | Scannable differences | Same criteria for every product |
| Who each option is for | Disambiguation | Explicit "choose A if…" |
| Pricing snapshot | Reduces hallucination | Numbers only if public and dated |
| Limitations | Trust | What your product does not do |
| Methodology | Experience signal | How you tested or what you based this on |
| Disclosure | Trust | Affiliate, employer, or partner relationships |
Skip fake 9.8/10 scores unless a real methodology sits behind them. Invented rubrics are theater.
Step-by-step: write a comparison assistants can use
- Pick a real decision. Two to four options a buyer actually debates — not twelve logos for SEO coverage.
- Lock criteria before prose. Example for analytics: data sources, time-to-value, seat pricing model, SSO, export limits.
- Build the table first. Fill cells with facts. If a cell is unknown, write "not published" — do not guess.
- Write "best for" blurbs of 2–3 sentences each. No insults; contrast constraints.
- Add a dated pricing note. "Public pricing as of July 2026; confirm on vendor sites." Link out.
- Disclose relationships near the top or directly under the verdict — employer bias, affiliate links, sponsored placements.
- Put the answer near the top. First screenful should state how to choose, not a history of the category.
- Mark up the page as Article/BlogPosting with a real author. Use Product schema on product pages, not as a fake review score on the comparison.
- Ensure HTML is visible without JS so crawlers see the table.
- Revisit quarterly or when plans change; update the date honestly.
Honesty beats affiliate sludge
Assistants and humans both punish pages that:
- Rank every competitor "poor" except the author’s employer.
- Hide pricing behind "book a demo" when public plans exist (or invent plan names — fix your own pricing clarity first).
- Use scraped screenshots without context.
- Stuff synonyms ("best alternative best competitor best vs").
If you sell one of the products, say so in the first screenful. "We build Northstar Analytics. Here is how we differ from SheetStack, including where they win." That sentence earns more trust than a neutral cosplay. It also supports E-E-A-T by making experience and bias visible.
Table example (hypothetical)
| Criterion | Northstar Analytics | SheetStack |
|---|---|---|
| Best for | Product-led B2B SaaS | Spreadsheet-heavy ops teams |
| Time-to-first dashboard | ~1 day with Segment | Same day if data already in Sheets |
| SSO | On Growth plan | Enterprise only |
| Public starting price | $49/mo (illustrative) | $29/mo (illustrative) |
| Weak spot | Limited offline exports | Weak multi-touch attribution |
Label illustrative numbers as hypothetical if they are not your real public prices. Never invent competitor prices.
Competitor pages you do not control
Third-party roundups on review sites will keep getting cited. Claim your profiles, fix categories, and stop fighting reviews you have not earned — covered in review sites AI keeps citing. Your owned comparison page should agree with those profiles on product name and positioning. Contradictions ("we are enterprise-only" on your site, "great for freelancers" on G2) give models mush to average.
Mistakes to avoid
- Publishing "vs" pages for products you have never used and citing no sources.
- Blocking AI crawlers while hoping for AI citations.
- Updating the title to "2026" without changing content — a freshness lie.
- Mixing parent company and product brand until the entity blurs — keep names distinct as in multi-brand entity separation.
- Building twenty thin vs-pages for long-tail SEO that disagree with each other.
How to verify
Fetch the URL with a text-only client. Confirm the table and disclosure appear. Check that Organization/Product names match the rest of the site. Spot-check two prompts in an assistant: if it misstates your plans, your page (or pricing page) is still ambiguous.
A BrandKnown scan will catch access and identity issues around the page; it will not promise that Perplexity will recommend you. Build the comparison worth citing, then measure referrals without over-reading small samples.