All 204 readable homepages in my 232-company benchmark let AI search in. Here is what the headlines miss, and why the fix is a differentiator your best-fit buyer can check.
Your buyers now meet you through a machine before they meet you on your site. ChatGPT, Perplexity, Claude and Google’s AI Overviews read your homepage, lift a sentence and decide what you are. If that sentence could belong to any vendor in your category, the engine has nothing to attribute to you.
I checked 232 B2B homepages to see how often that happens. In 157 of 169 headlines I rated, the page says nothing only that company could say. A competitor in the same category could put the headline on its own homepage unchanged. One in four would work for a company in a different industry.
The cost lands in two places marketing leaders care about.
- Authority. An answer engine needs a sentence it can attribute. A headline with no category, no buyer and no proof gives it nothing to cite, so it cites the brand it already knows.
- Authenticity. Buyers have read years of AI-written copy. “The platform for modern teams” now reads as generated, whether a person wrote it or not. Sameness costs you trust before a rep ever gets on a call.
AI makes this easy to get wrong. Teams use the same models, trained on the same category copy, to write their homepages. Left alone, the models return the average of the category. Every vendor that ships the first draft sounds like every other vendor that shipped the first draft.
The technical side is already handled. All 204 readable homepages in this sample let every AI search crawler in. No file, plugin or tool will fix the words on the page.
The fix is a decision your team makes once and then holds. Name what you sell. Name the buyer you serve best. State one fact about you that a competitor cannot copy. Put that on the homepage and in every field a machine reads.
This report shows where the homepages in this sample fall short of that standard and what to change, section by section.
Download the full AI Sameness Report (PDF)
| Rating | Headlines |
|---|---|
| Swappable | 157 |
| Only this company could say it | 12 |
What I measured
I ran one fixed set of checks on each homepage. Every number in this report is one of two kinds.
Counted numbers come straight from the page source. You can check any of them yourself in a browser.
Judged numbers come from the headline test. Two separate runs of the same AI model scored each headline against one fixed rubric. A rating counts only when both runs agree. They gave the same swap rating to 155 of 169 headlines (92%). A second AI model from a different company then rated every headline blind. It matched the first two on 88% and 87% of swap ratings, and the headline figures held.
Most of the sample sells B2B software. The full method and its limits are at the end.
Is anything blocking AI search?
No. I ran five technical checks, and the basics are in place across the sample.
| Check | Sites passing of 204 |
|---|---|
| robots.txt allows all six AI search crawlers | 204 |
| Homepage text present before JavaScript runs | 203 |
| No directive blocking indexing or snippets | 203 |
| Real 404s, working sitemap and canonical tag | 170 |
| Company schema on the homepage | 145 |
| Check | Sites passing |
|---|---|
| robots.txt allows all six AI search crawlers | 204 of 204 |
| Homepage text is present before JavaScript runs | 203 of 204 |
| No directive blocking indexing or snippets | 203 of 204 |
| Real 404s, a working sitemap, and a canonical tag | 170 of 204 (83%) |
| Company schema on the homepage | 145 of 204 (71%) |
Zero sites block an AI search crawler. Nine sites block a crawler that collects training data. That is a business decision about model training. It does not remove you from AI search answers.
One site in the sample carries a noindex tag on its homepage. That tag asks search engines to drop the page. Google states that a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode. One line of code takes the homepage out of both.
Why this matters
A plumbing failure blocks everything downstream. A blocked crawler hides the page. So does a noindex tag. So does a homepage built entirely in JavaScript. Any one of them makes every word invisible to some or all of the engines. That is why vendors lead with it.
This group already cleared that bar. If your team spends next quarter on crawler access, it spends the quarter on a problem it does not have.
What to do
- Check the three hygiene items once: a real 404 page, a sitemap that loads, and a canonical tag on the homepage. 17% of sites miss at least one.
- Add company schema if you lack it, with links to your LinkedIn page and other profiles. 29% of sites have none. It tells a machine which company owns the site.
- Search your homepage source for “noindex”. It takes ten seconds.
- Put this check on a quarterly calendar. Do not buy a tool for it.
The file nobody asked for
87 of 204 sites publish an llms.txt file. That is 43%.
| Group | Sites |
|---|---|
| Publish llms.txt | 87 |
| Of those, block training crawler | 4 |
| Of those, no company schema | 13 |
| Headlines that say what they sell, llms.txt sites | 60% |
| Headlines that say what they sell, sites without | 41% |
The file is a summary written for AI systems. No major answer engine says it reads it. Google says you don’t need AI text files like llms.txt to appear in AI Overviews or AI Mode. Several of the 87 never chose to publish one. An SEO plugin generated the file by default.
Four sites publish llms.txt and block an AI training crawler in the same robots.txt. They invite the machine in one file and turn it away in the next. Thirteen of the 87 have no company schema. They shipped the new file before the old basics.
The publishers are not careless teams. 60% of their headlines say what the company sells, against 41% for sites without the file. They pay attention to AI search. The attention went to the plumbing.
Why this matters
43% shows where marketing attention went. A team that writes, debates and publishes an llms.txt file spent real time on a file the engines skip. That time came from somewhere. The homepage headline is a better place to spend it.
What to do
- Stop spending time on llms.txt. You do not need to delete a file that already exists.
- If a plugin generated it, leave it alone.
- Move those hours to the homepage headline and description. The next two sections show where.
The swappable headline
A swappable headline is one that another company could put on its own homepage without changing a word.
- Any industry could use it 44 26%
- Any category competitor could use it 105 62%
- Raters disagreed 14 8%
- Only this company 6 4%
| Rating | Headlines of 169 |
|---|---|
| Any industry could use it | 44 |
| Any category competitor could use it | 105 |
| Raters disagreed | 14 |
| Only this company | 6 |
| Says what | 84 |
| Says who | 21 |
| Specific detail | 17 |
| Neither | 69 |
Two raters judged each headline on its words alone. They did not credit the line beneath it or the page title. Both accepted 169 headlines as real positioning statements. Here is how those 169 scored.
| Rating | Headlines |
|---|---|
| A company in any industry could use it unchanged | 44 of 169 (26%) |
| Any competitor in the category could use it | 105 of 169 (62%) |
| The two raters disagreed | 14 of 169 (8%) |
| Only this company could say it | 6 of 169 (4%) |
The title figure adds the first two rows to the eight split headlines where neither rater picked “only this company”. That makes 157 of 169 (93%) with no “unique” vote. These eight are the one place the report counts a split rating. Neither run called them unique. The second model’s ratings leave the count at 157.
Read the first row again. One in four headlines would work for a company in a different industry. Here are eight of them. I removed the names.
- Be found
- Simply resilient
- Your edge starts here
- Build faster. Scale smarter.
- Say yes to more.
- When numbers matter
- Next gen results
- Your future workforce
You cannot name the industry from any of them. An answer engine cannot either.
The headlines also fail a stranger’s two basic questions. Only half, 84 of 169, say what the company sells. Only 21 say who it is for. Only 17 contain a detail a reader could check, such as a number or a named product. 69 answer neither question.
The company name is missing too. Of the 187 sites with a main headline tag, only 17 put the company name in it.
Why this matters
An answer engine does not read your homepage the way a buyer does. It does not see the logo, the design or the demo video. It lifts sentences, decides what they claim, and decides who to attribute them to. A headline with no company name, no category and no buyer gives it nothing to lift.
A study by geoSurge, which sells AI-visibility software, looked at which brands AI models search for. The models searched for brands they already knew 55.7% of the time. For brands outside their top 10, the figure was 17.4%. The authors say this shows a link, not proof that memory causes the searches.
Familiarity is exactly what a swappable headline fails to build. When every vendor in a category says “the platform for modern teams,” the engine falls back on the names it already knows. The incumbent wins by default.
The cost does not stop at AI search. Your homepage headline is the source text for the rest of your marketing. Sales reps paste it into outbound emails. Partners paste it into co-marketing pages. New hires learn the company from it. Agencies brief from it. If the line is swappable, every copy of it is swappable too.
A buyer comparing three vendors with interchangeable headlines has one way left to choose. Price.
What to do
Run the swap test on your own headline before you rewrite anything.
- Copy your headline into a document.
- Put your closest competitor’s name in front of it.
- Read it aloud. If the sentence is still true, it is swappable.
- Put a company from a different industry in front of it. If it still works, the headline says nothing about what you sell.
Then rewrite the headline to answer three questions: what you sell, which buyer it is for, and one differentiator that buyer could check. If you draft with AI, give it your differentiators and your ICP first. Otherwise it writes the category average.
Here is the pattern on an invented company, Fieldline, which sells backup software. It is an illustration, not a sampled site.
| Headline | |
|---|---|
| Before | Built for what’s next. |
| After | Backup and recovery for mid-market IT teams. Restore a full server in under 15 minutes. |
The after version is longer. It names the category, the buyer and a measurable claim. A competitor cannot copy it without also copying the proof. An answer engine can lift either sentence and attach it to Fieldline.
Ship a line your competitors cannot borrow. Sharpen it later.
Why does a short headline say less?
Short headlines are the most swappable headlines in the sample.
| Group | Count | Base |
|---|---|---|
| ≤4 words | 16 | 28 |
| 5+ words | 28 | 141 |
| Stacked fragments | 11 | 25 |
| Single sentence | 33 | 144 |
| Mentions AI: says what | 38 | 53 |
| No AI: says what | 46 | 116 |
| Headline type | Rated usable by any industry |
|---|---|
| Four words or fewer | 16 of 28 (57%) |
| Five words or more | 28 of 141 (20%) |
| Stacked fragments (“Any cloud. One platform.”) | 11 of 25 (44%) |
| A single sentence or phrase | 33 of 144 (23%) |
The punchy three-word headline is the style most teams reach for. It reads well in a design review. It also tells a stranger the least. A four-word headline has no room for a category and a buyer, so it settles for a mood.
Stacked fragments have the same problem. Three short claims in a row sound confident. Each one could belong to anyone.
I expected the AI headlines to be the empty ones. They are not. “AI” appears in 36 of 187 headlines, more than twice as often as any other word. Half of all sites mention AI or agents somewhere near the top of the homepage. Yet the headlines that mention AI say what the company sells 72% of the time. The headlines that skip it manage 40%.
Naming the technology forces the writer to name the product. The vague headlines in this sample are the ones that avoided saying anything concrete at all.
The smaller groups hold 25 to 53 headlines each. Read the gaps as direction, not as precise measures.
What to do
- Give the headline room. Eight to fourteen words is enough for a category, a buyer and a claim.
- If design insists on a short headline, make the line directly beneath it do the work. Put the category and buyer there in plain words.
- Replace stacked fragments with one sentence that makes one claim.
- Name the thing you sell. If it uses AI, say what the AI does and for whom.
Four descriptions of one company
Your homepage describes you in several places a visitor never sees. The search description shows in Google results. The share-card description shows when someone posts your link. The structured data tells machines what the company is. Those fields should say the same thing.
| Item | Sites |
|---|---|
| 0 | 5 |
| 1 | 114 |
| 2 | 69 |
| 3 | 14 |
| 4 | 2 |
| Meta description never states the company name | 74 of 194 |
| Share-card title differs from page title | 48 of 204 |
| Share-card title starts with “Home” | 24 of 204 |
| Main headline tag holds a badge, promo or label | 18 of 187 |
| No main headline tag at all | 16 of 204 |
On 85 of 204 sites, they do not. Sixteen sites describe themselves three or more ways. At least one homepage in the sample carries four different statements of what it sells, counting its headline.
The smaller slips are easier to see.
| Slip | Sites |
|---|---|
| Meta description never states the company name | 74 of 194 with a meta description (38%) |
| Share-card title differs from the page title | 48 of 204 (24%) |
| Share-card title starts with the word “Home” | 24 of 204 (12%) |
| Main headline tag holds a badge, promo or page label | 18 of 187 with a headline tag (10%) |
| No main headline tag at all | 16 of 204 (8%) |
The share-card title is what LinkedIn and Slack show when someone posts your homepage. On 24 sites, that title reads “Home” or “Homepage”. Google also names the share-card title as one of the sources it uses to write the title link in search results.
WordPress sites show the “Home” share card most. It appears on 13 of the 48 WordPress sites (27%), against 11 of the other 156 (7%). A default setting wrote that title, and nobody changed it.
The headline tag slip matters for a different reason. On 18 sites, the most prominent tag on the page holds an analyst badge, an award, an event promotion or a page label. One holds a mortgage license number. A machine looking for the company’s main claim finds a press release.
Why this matters
Inconsistent descriptions tell a machine you are four different things. An engine has no way to know which one you meant. Your sales team has the same problem. When the homepage, the search result and the share card disagree, the buyer gets three first impressions and keeps none of them.
These are the cheapest fixes in this report. Each one is a single field in your CMS. Nobody needs a budget.
What to do
- Write one approved company description in two lengths: 25 words, and 155 characters.
- Put the 155-character version in the meta description, the share-card description and the structured data. Use the same words in all three.
- Start that description with the company name.
- Set the share-card title to your company name plus the category. Never “Home”.
- Give the main headline tag to your positioning line. Move badges and event promotions to their own section.
One person can do all five in an afternoon.
Why does a differentiator compound?
Crawler access decides whether a machine can read your homepage. Your differentiators decide what it repeats, and to whom.
A real differentiator compounds. A headline built on a real differentiator for a named buyer gives answer engines a sentence to quote and attribute. It gives your sales team a line to repeat. It gives your content team a brief to write against. Every blog post, product page and email your team ships either repeats one claim or adds a fifth description to the pile.
The teams in this sample are not short on skill. They are senior marketing leaders, and all 204 let the crawlers in. Differentiation is harder because it requires a decision. Someone has to choose what you sell, which buyer you serve best and what you can prove to that buyer. Then the team has to hold that line in every channel.
No tool makes that decision for you. An AI writing tool left to itself makes the opposite one. It writes the category average.
A 30-day fix list
Here is the order I would work in.
- Week 1Plumbing, once
- Search source for “noindex”; check 404, sitemap, canonical
- Share-card title = company name + category
- One company description (25 words, 155 chars) in all three fields
- Week 2Decide the differentiator
- Run the swap test with marketing and two sales reps
- Start hereList top three differentiators for your best-fit buyer, each with proof
- Draft three headlines: category, buyer, one checkable differentiator
- Week 3Test with buyers
- Test the three drafts with five buyers or five reps
- Week 4Ship and match
- Ship the winner; update decks, signatures and partner pages
| Week | Fix | Owner |
|---|---|---|
| 1 | Search the homepage source for “noindex” and check the 404 page, sitemap and canonical tag | Web lead |
| 1 | Set the share-card title to company name plus category | Web lead |
| 1 | Write one company description in 25 words and 155 characters, and put it in all three fields | Head of product marketing |
| 2 | Run the swap test on the headline with your marketing team and two sales reps | VP Marketing |
| 2 | List your top three differentiators for your best-fit buyer, each with proof | Product marketing |
| 2 | Draft three replacement headlines that name the category, the buyer and one checkable differentiator | Product marketing |
| 3 | Test the three drafts with five buyers or five reps who talk to buyers | Product marketing |
| 4 | Ship the winning headline and update sales decks, email signatures and partner pages to match | VP Marketing |
Skip the llms.txt debate. Skip the new tool. Put the hours into weeks two and three.
How I ran this
The sample is 232 companies where a VP-level or higher marketing leader is a LinkedIn connection of mine. It leans toward B2B software, with some services, finance, healthcare and education. It is a convenience sample, not a random sample of any market.
I checked one page per company, the homepage, on one day. A script fetched each page as a browser and as each of six AI search crawlers. It also read robots.txt, the sitemap, a deliberately broken URL and the page source. I could read 204 sites. Another 26 refused a request from a cloud address, and 2 did not respond. Those 28 are unscored, and they skew toward large enterprises.
The judged numbers have three limits.
- The two main rater runs come from the same AI model using one rubric. So I had a model from a different company rate all 187 headlines without seeing their answers. Under a two-of-three vote, 163 of 169 headlines (96%) said nothing only that company could say. Counting a headline as unique if any one of the three raters thought so, the figure is 157 of 169 (93%). I report the 93%. Agreement across models shows the rubric travels. It does not prove the rubric is right.
- The test is strict. Only the headline counts, and many sites say what they sell in the line beneath it.
- The raters judged wording alone. They did not compare each headline against competitors’ actual sites.
One more number stays out of the tables. Thirteen sites refused a request that carried an AI crawler’s name. That could be blocking. It could also be protection against impostors, and I cannot tell the two apart from outside.
Run the test on your own homepage
It takes one minute.
- Paste your homepage link into a LinkedIn post draft. Read the card title.
- Read your headline to someone outside your industry. Ask what you sell.
- Put a competitor’s name in front of your headline. See if it still works.
If your company is one of the 232, I will send you your row. If it is not, send me your homepage link and I will run the same check.
Before you hire an agency or buy an AI visibility tool, ask one question. If an answer engine repeated our headline with no logo beside it, would anyone know it was us?
Scott King helps B2B technology companies build AI marketing systems and create differentiated content.