Comparison tables and buying guides AI assistants can use

Many shopping questions put to AI assistants come down to "which one should I buy?". Comparison tables and buying guides are where you answer that question yourself, in your own words, with your own products. This guide covers how to build tables agents can read, how to keep attributes comparable, and how to write buying guides that are honest enough to be quoted.

Why comparison content matters to AI agents

A shopper asks: "Which Northwind grinder should I get for pour-over, under $200?" The assistant has to line up the options, find the attributes that matter (grind range, burr type, price) and give a reason. It can do that the hard way, by opening every product page and pulling facts out of each one. Or it can find a page where you've already done the work.

A good comparison table hands over the facts side by side. A good buying guide hands over the reasoning: which product suits which person, and why. Both make your store easier to recommend, because the assistant can back up its answer with your page. For the wider picture of how assistants pick, see how AI assistants choose products and stores.

1. Shopper asks "Which grinder for pour-over, under $200?"
2. Agent needs facts Grind range, burr type, price
3. Finds your table All three products, same rows, same units
4. Finds your guide "Best for pour-over" with a reason

Build tables as real HTML tables

Many comparison charts look like tables but aren't. Some are a single image. Some are a grid of styled boxes. Some only appear once a script runs, or sit in a carousel that shows three columns at a time. People can read all of these. Agents often can't.

• An image of a table has no text at all, unless someone wrote the whole table into its alt text, which nobody does.
• A grid of boxes (div elements styled to line up) has the text, but nothing says which value belongs to which product and which attribute. The relationship only exists visually.
• A table built by JavaScript is missing for agents that read the HTML without running scripts. See why AI agents can't see your JavaScript-only content.

A real HTML table states the relationships in the markup. The W3C's tables tutorial puts it simply: header cells are marked up with <th> and data cells with <td>, and scope says whether a header applies to a row or a column. That's exactly what an agent needs to know that "Steel flat" describes the burrs of the Pro.

<table>
  <caption>Northwind grinders compared</caption>
  <thead>
    <tr><th scope="col"></th>
      <th scope="col">Basic</th>
      <th scope="col">Plus</th>
      <th scope="col">Pro</th></tr>
  </thead>
  <tbody>
    <tr><th scope="row">Price</th>
      <td>$89</td><td>$149</td><td>$229</td></tr>
    <tr><th scope="row">Burrs</th>
      <td>Ceramic conical</td><td>Steel conical</td><td>Steel flat</td></tr>
    <tr><th scope="row">Grind settings</th>
      <td>18</td><td>40</td><td>Stepless</td></tr>
  </tbody>
</table>

The <caption> names the table, so an agent knows what's being compared without reading the surrounding page. On phones, let the table scroll sideways inside its container rather than rebuilding it as stacked cards in a script. The HTML stays the same for every reader.

Looks like a table

• An image of the chart
• Boxes lined up with CSS
• Columns built by a script
• Ticks drawn as icons with no text

Is a table

• <table> with a caption
• <th> for products and attributes
• In the HTML from the server
• "Yes" and "No" in words

Make every row comparable

A table is only as useful as its consistency. The same rules that make product descriptions comparable apply, but in a table any slip shows at once:

• Same attributes for every product. If one column has no value for "Weight", write "Not stated" or find the number. An empty cell reads as "doesn't apply".
• Same units in every cell. Don't mix grams and ounces across columns. Put the unit in the cell ("410 g"), or in the row header ("Weight (g)") and nowhere else.
• Words, not only symbols. A green tick icon with no text means nothing to an agent. Write "Yes" (the icon can stay beside it) or give the icon a text alternative.
• Specific values, not ratings. "Battery: 14 hours" beats "Battery: ★★★★". Stars and bars are your judgment; numbers can be checked.
• The same facts as the product pages. If the table says 14 hours and the product page says 12, an assistant has to choose. Pull both from the same product data where you can.

Keep prices in tables current or leave them out. A comparison page that was right at launch and wrong six months later is worse than one that links to each product for the price.

Write buying guides that answer "which should I buy?"

A buying guide is where you explain the choice, not just list the options. Write it the way a good shop assistant talks:

• Organize by situation. Headings such as "For pour-over at home", "For espresso" and "For travel" match the way people phrase their questions. See titles, descriptions and headings AI agents understand.
• Give a clear pick with a reason. "For pour-over, choose the Plus: steel burrs give a more even medium grind than the Basic, and you don't need the Pro's espresso range." An assistant can quote that sentence whole.
• Say what decides it. Explain the two or three attributes that matter for the choice, in plain words, so the guide is useful even for products you don't sell.
• Link each pick to its product page with descriptive link text, so an agent can go straight to price, stock and add to cart.
• Keep it in HTML text. Not in a video, a PDF or a quiz that only gives an answer after five clicks. If you have those, put the conclusions on the page too. See PDFs and documents AI agents can read.

northwind.example/guides/choosing-a-grinder

For pour-over at home Heading matches the question

Choose the Plus: steel burrs give an even medium grind. A pick and a reason

Not for espresso: its finest setting is too coarse. Honest limit

See the Northwind Plus

Be honest about trade-offs

A guide where every product is "perfect for everyone" gives an assistant nothing to work with. It needs to know who each product is not for, so it can match the right one to the shopper's situation.

• Name the limits. "The Basic is loud and slow for more than two cups." That sentence helps the shopper who makes one cup a day choose it with confidence.
• Recommend the cheaper option when it's enough. It builds trust with shoppers and gives the assistant a reason to treat your guide as fair.
• Agree with your reviews. If customers say the Plus is hard to clean, a guide that calls it "effortless to maintain" contradicts evidence the assistant can also read. See reviews and ratings AI agents can read.
• Date it and update it. Put "Updated October 2026" at the top and revise when products change. A guide that recommends a discontinued model sends agents to a dead end.

An assistant needs to know who each product is not for, so it can match the right one to the shopper.

What AgentScore checks

AgentScore doesn't look for comparison tables or buying guides. Its Readability check "Content loads without JavaScript" (4 points) covers the groundwork: it compares how much of your home page's text is in the HTML before JavaScript runs with how much appears after, and passes when at least 70% is already there. It runs on the home page only, so a comparison chart built by a script on another page won't show up in it. Use the same test by hand on your guides: open one with JavaScript turned off and see whether the table and the picks are still there.

Where to start