How ready are software company websites for AI search? What a check of 1,586 sites showed

The median software company website scores 68 out of 100 for AI search readiness, and only 21.8% reach the 80-plus band my checker labels "Ready to be cited". Getting in is rarely the problem: Access averaged 93.1% of its points and caused 5.0% of all points lost. What is missing is content an engine can quote. Answerability and Evidence together account for 76% of the points lost, led by question-shaped headings, answer-first sections, sourced facts, comparison tables and FAQ sections. I ran 1,601 software company websites through version 2 of my AI search readiness check on 22 October 2026 and scored 1,586 of them: software publishers with 11 to 200 staff, founded 2015 or later, headquartered in the US and 12 European countries.

By Ronan Liedmeier, fractional head of growth. Last updated 23 October 2026.

An AI search readiness score is a 0 to 100 estimate of how ready a site is to be read and quoted by AI search engines such as ChatGPT search, Perplexity and Google AI Overviews. Mine is built from 22 checks a crawler can run, in four groups: Access, Answerability, Evidence and freshness, and Entity and trust. It measures readiness. Whether a site actually gets cited is a separate question, which I tested in the 209-vendor study.

What do AI readiness scores look like for software company websites?

Most software company websites land in the middle: the median score was 68 and the mean 63.9 across 1,586 sites. Half scored between 54 and 78, the 10th percentile was 37 and the 90th was 87, with a minimum of 4 and a maximum of 98.

The distribution leans high with a long tail at the bottom. 92 sites (5.8%) scored 90 or more. 333 sites (21.0%) scored under 50, and many of the lowest scores belong to the 105 sites whose homepage did not load for the checker, which had a median score of 13 (more on that below).

The checker gives every result one of four verdicts. Summed by band:

The typical software company website sits in the third band.

Where do software company websites lose points?

They lose them in content. Answerability accounted for 46.7% of all points lost and Evidence and freshness for 29.5%, together 76% of the 51,121 points lost across the sample, while Access accounted for 5.0%.

The category averages show the same thing from the other side. Access averaged 93.1% of its applicable points (median 100%). Entity and trust averaged 62.5% (median 64.7%). Evidence and freshness averaged 52.2% (median 56%). Answerability averaged 47.5% (median 50%), so the typical site earned half of the points for content an engine could answer from.

Ranked by total points lost, the top ten checks:

The access checks sit at the bottom of the full ranking: text without JavaScript 1,242 points, possible bot blocking 636, HTTPS homepage 420, noindex 244 and robots.txt 32.

Points lost is not the same as importance for citation. Question-shaped headings and answer-first sections cost the most points, but in the 209-vendor test answer-first paragraphs showed no relationship with being cited. Comparison tables and FAQ sections cost fewer points and did predict citation.

Can AI crawlers get into software company websites?

Mostly yes. One in eight software company websites has a hard access problem, and almost none of them is caused by robots.txt.

A hard access problem here means the homepage did not load over HTTPS, the pages had too little text without JavaScript, robots.txt blocked a search bot, or the homepage carried noindex or nosnippet. 201 sites (12.7%) had at least one. Counting softer issues too, such as a partial JavaScript pass or possible bot blocking, 391 sites (24.7%) had some access problem.

The robots.txt finding surprised me most. Only 4 sites disallowed OAI-SearchBot, PerplexityBot, Googlebot or Bingbot, and 1,576 (99.4%) had no robots.txt problem. The check scores the search bots only, because they are the ones that decide whether you appear in answers. OpenAI's documentation says sites that disallow OAI-SearchBot "will not be shown in ChatGPT search answers", while disallowing GPTBot only signals that content should not be used for training (OpenAI). Perplexity says the same of PerplexityBot, which "is not used to crawl content for AI foundation models" (Perplexity).

The bigger problems are JavaScript and homepages that do not load. 170 sites (10.7%) failed the text-without-JavaScript check, and on the 1,481 sites whose homepage loaded the figure was 87 (5.9%). That matters because, in Vercel's crawler analysis, "none of the major AI crawlers currently render JavaScript", with Google's Gemini and AppleBot as the exceptions (Vercel). The 105 sites whose homepage did not load averaged 21.7 (median 13), against 66.9 (median 69) for the 1,481 that did. Many of those block automated requests in general, so their low score mostly reflects that block rather than their content.

160 sites (10.1%) were flagged for possible bot blocking: 108 refused or short-changed only the bot user agents, and 52 refused the browser-like request too. On the 1,481 reachable sites, 105 (7.1%) showed possible bot-only blocking. My check sends each bot's user agent from a residential connection, not from the bots' own IP ranges. Cloudflare verifies bots through cryptographic signatures or published IP lists (Cloudflare), so a firewall that does this can refuse my fake bot and still let the real one through. Treat these as possible blocks to verify in your server logs.

The detail is in two sibling pieces: do software companies block AI crawlers? and software websites without JavaScript.

Which content formats do software company websites use?

Few of them. Of the 1,303 sites where the check found article pages, 27.9% passed none of the four answer formats I check (question headings, answer-first sections, a comparison table, an FAQ), and only 4.1% passed all four.

The 1,303 is the right base for this section. The other 283 sites (17.8%) had no article page the check could find, so the format checks fail on them automatically. Those 283 averaged 32.1 (median 36), against 70.8 (median 72) for sites with articles.

The spread of the four formats per site, at pass level:

The mean was 1.37 formats per site. 338 sites (25.9%) had both an FAQ and a comparison table, the two formats in this group that predicted citation in the 209-vendor test. The score rises with each format by construction, since each one is a scored check, so the mean score column describes the sites rather than proving anything new.

Many articles are also short and unsourced. 255 sites (19.6%) had a median article under 400 words, and 568 (43.6%) had a median of less than one outbound link in article content. The full breakdown is in which content formats AI engines quote.

Can AI search tell who the company is and what it costs?

Partly. Fewer than half of software company websites have complete Organization structured data or three linked profiles, and only 22.7% of commercial sites state prices on a pricing page.

Pricing uses a base of 1,361 commercial sites, leaving out 225 that did not appear to sell something with a list price. 309 (22.7%) had a pricing page with amounts, 791 (58.1%) had a pricing page without amounts or prices elsewhere, and 261 (19.2%) had neither.

These checks help an engine identify the company and answer a buyer's cost question from the company's own site. In the 209-vendor test, neither Organization schema nor sameAs links had a significant relationship with being cited. Ahrefs found the same in a test of 1,885 pages that added JSON-LD: "Adding schema produced no major uplift in citations on any platform" (Ahrefs). More in can AI search tell who your company is?

Are US software company websites more ready than European ones?

On the overall score, about the same. The mean score was 64.9 for the 639 US sites and 63.3 for the 947 European sites, and the 1.6-point gap is not statistically meaningful (z 1.59).

Three specific gaps are real: FAQ sections, comparison tables and prices on the pricing page.

Two of the three real gaps, FAQ and tables, are formats that predicted citation in the 209-vendor test. A European software company with an FAQ and a comparison table on its articles is ahead of most of its European peers on the two cheapest features to add. I am not ranking individual countries: most have under 100 sites in the sample, and the regional differences in mean score are small.

How was the study run, and what are its limits?

I checked every site once with version 2 of the GMS AI search readiness check, in one continuous run on 22 October 2026 between 18:03 and 18:28 Pacific time. The biggest limit is that the score measures readiness, and the overall score did not predict citation when pooled in the 209-vendor test. That is the same scoring code as the free tool, run as a local batch on a residential connection.

The score is points earned divided by points applicable, times 100. Checks that do not apply leave the denominator: a site with no article pages, for example, is not scored on article depth.

The limits, in order of how much they matter:

1. Readiness is not citation. The 209-vendor study found the overall score did not predict citation when pooled across vendors. Five checks did within the same answer: FAQ, named author with about page, article depth, comparison tables and visible dates. Organization schema, sameAs, answer-first paragraphs and recency did not.
2. Residential connection. Sites that treat residential traffic differently from the edge network the live tool uses can score differently here. The 105 homepages that did not load are the clearest case.
3. The bot-blocking check is a user-agent test, not a test from the bots' IP ranges, so it reports possible blocking only.
4. The sample is software companies with 11 to 200 staff, founded 2015 or later. It says nothing about larger software companies or other industries.
5. Country is the HQ filter of the company search, not verified company by company.
6. One check per site, one day. A site mid-redesign or briefly down on 22 October is scored as it was that evening.

What I'd do

Ordered by points lost and by what the 209-vendor test found predicts citation. Steps 1 and 2 are prerequisites; steps 3 to 5 are the five checks with citation evidence; steps 6 and 7 are score work without it.

1. Fix access first. Check that the homepage loads over HTTPS, that the main text is in the raw HTML, and that robots.txt allows OAI-SearchBot and PerplexityBot. 12.7% of the sample had a hard problem here, and nothing else counts until it is fixed.
2. If the site has no articles, start there. The 283 sites without an article page averaged 32.1 against 70.8 for sites with articles.
3. Add an FAQ and a comparison table to the main articles. They cost the sample 3,640 and 4,604 points, both predicted citation, and only 25.9% of article sites had both.
4. Put a named author with an about page and a visible date on every article. 49.8% passed the author check and 61.6% the date check, of the 1,303 article sites.
5. Write articles long enough to be worth lifting. 19.6% of article sites had a median article under 400 words; the depth check passes at 800.
6. Then work on question headings, answer-first sections, facts and outbound sources. They are the four biggest point losses in the sample, but the 209-vendor test found no citation effect for answer-first paragraphs and only a weak one for outbound sources. Do them because they make pages easier to read and lift. They are not proven to get you cited.
7. Leave Organization schema, profile links and brand consistency for last. They help an engine identify you, and the 209-vendor test found no citation effect for schema. State prices on the pricing page while you are there: only 22.7% of the 1,361 commercial sites did.

If you run growth at a software company and want to know where your own site sits in this distribution, run the free readiness check: it is the same scoring code used for this study. If you want the citation side as well, the AI Visibility Snapshot covers it: ten prompts in your category, your site scored on the 22 checks, and a one-page readout in three working days. For the rest of the method, see how to get cited by ChatGPT, Perplexity and Google AI Overviews.

Re-check by 23 January 2027: re-run the same 1,601 domains with the same code to see whether the median of 68 and the format pass rates move, re-test the possible bot blocks from a second network, and update the citation column once the second 209-vendor run with ChatGPT and Google AI Overviews is in.